diff --git a/CHANGELOG.md b/CHANGELOG.md new file mode 100644 index 0000000000000000000000000000000000000000..c45c651d60ef0dde5b7b5ddba82db5246f32f7a0 --- /dev/null +++ b/CHANGELOG.md @@ -0,0 +1,200 @@ +## 1.3.2 + +### Bug Fixes: + * fix files served out of tmp directory even if they are saved to disk + * fix postprocessing overwriting parameters + +## 1.3.1 + +### Features: + * revert default cross attention optimization to Doggettx + +### Bug Fixes: + * fix bug: LoRA don't apply on dropdown list sd_lora + * fix png info always added even if setting is not enabled + * fix some fields not applying in xyz plot + * fix "hires. fix" prompt sharing same labels with txt2img_prompt + * fix lora hashes not being added properly to infotex if there is only one lora + * fix --use-cpu failing to work properly at startup + * make --disable-opt-split-attention command line option work again + +## 1.3.0 + +### Features: + * add UI to edit defaults + * token merging (via dbolya/tomesd) + * settings tab rework: add a lot of additional explanations and links + * load extensions' Git metadata in parallel to loading the main program to save a ton of time during startup + * update extensions table: show branch, show date in separate column, and show version from tags if available + * TAESD - another option for cheap live previews + * allow choosing sampler and prompts for second pass of hires fix - hidden by default, enabled in settings + * calculate hashes for Lora + * add lora hashes to infotext + * when pasting infotext, use infotext's lora hashes to find local loras for `` entries whose hashes match loras the user has + * select cross attention optimization from UI + +### Minor: + * bump Gradio to 3.31.0 + * bump PyTorch to 2.0.1 for macOS and Linux AMD + * allow setting defaults for elements in extensions' tabs + * allow selecting file type for live previews + * show "Loading..." for extra networks when displaying for the first time + * suppress ENSD infotext for samplers that don't use it + * clientside optimizations + * add options to show/hide hidden files and dirs in extra networks, and to not list models/files in hidden directories + * allow whitespace in styles.csv + * add option to reorder tabs + * move some functionality (swap resolution and set seed to -1) to client + * option to specify editor height for img2img + * button to copy image resolution into img2img width/height sliders + * switch from pyngrok to ngrok-py + * lazy-load images in extra networks UI + * set "Navigate image viewer with gamepad" option to false by default, by request + * change upscalers to download models into user-specified directory (from commandline args) rather than the default models/<...> + * allow hiding buttons in ui-config.json + +### Extensions: + * add /sdapi/v1/script-info api + * use Ruff to lint Python code + * use ESlint to lint Javascript code + * add/modify CFG callbacks for Self-Attention Guidance extension + * add command and endpoint for graceful server stopping + * add some locals (prompts/seeds/etc) from processing function into the Processing class as fields + * rework quoting for infotext items that have commas in them to use JSON (should be backwards compatible except for cases where it didn't work previously) + * add /sdapi/v1/refresh-loras api checkpoint post request + * tests overhaul + +### Bug Fixes: + * fix an issue preventing the program from starting if the user specifies a bad Gradio theme + * fix broken prompts from file script + * fix symlink scanning for extra networks + * fix --data-dir ignored when launching via webui-user.bat COMMANDLINE_ARGS + * allow web UI to be ran fully offline + * fix inability to run with --freeze-settings + * fix inability to merge checkpoint without adding metadata + * fix extra networks' save preview image not adding infotext for jpeg/webm + * remove blinking effect from text in hires fix and scale resolution preview + * make links to `http://<...>.git` extensions work in the extension tab + * fix bug with webui hanging at startup due to hanging git process + + +## 1.2.1 + +### Features: + * add an option to always refer to LoRA by filenames + +### Bug Fixes: + * never refer to LoRA by an alias if multiple LoRAs have same alias or the alias is called none + * fix upscalers disappearing after the user reloads UI + * allow bf16 in safe unpickler (resolves problems with loading some LoRAs) + * allow web UI to be ran fully offline + * fix localizations not working + * fix error for LoRAs: `'LatentDiffusion' object has no attribute 'lora_layer_mapping'` + +## 1.2.0 + +### Features: + * do not wait for Stable Diffusion model to load at startup + * add filename patterns: `[denoising]` + * directory hiding for extra networks: dirs starting with `.` will hide their cards on extra network tabs unless specifically searched for + * LoRA: for the `<...>` text in prompt, use name of LoRA that is in the metdata of the file, if present, instead of filename (both can be used to activate LoRA) + * LoRA: read infotext params from kohya-ss's extension parameters if they are present and if his extension is not active + * LoRA: fix some LoRAs not working (ones that have 3x3 convolution layer) + * LoRA: add an option to use old method of applying LoRAs (producing same results as with kohya-ss) + * add version to infotext, footer and console output when starting + * add links to wiki for filename pattern settings + * add extended info for quicksettings setting and use multiselect input instead of a text field + +### Minor: + * bump Gradio to 3.29.0 + * bump PyTorch to 2.0.1 + * `--subpath` option for gradio for use with reverse proxy + * Linux/macOS: use existing virtualenv if already active (the VIRTUAL_ENV environment variable) + * do not apply localizations if there are none (possible frontend optimization) + * add extra `None` option for VAE in XYZ plot + * print error to console when batch processing in img2img fails + * create HTML for extra network pages only on demand + * allow directories starting with `.` to still list their models for LoRA, checkpoints, etc + * put infotext options into their own category in settings tab + * do not show licenses page when user selects Show all pages in settings + +### Extensions: + * tooltip localization support + * add API method to get LoRA models with prompt + +### Bug Fixes: + * re-add `/docs` endpoint + * fix gamepad navigation + * make the lightbox fullscreen image function properly + * fix squished thumbnails in extras tab + * keep "search" filter for extra networks when user refreshes the tab (previously it showed everthing after you refreshed) + * fix webui showing the same image if you configure the generation to always save results into same file + * fix bug with upscalers not working properly + * fix MPS on PyTorch 2.0.1, Intel Macs + * make it so that custom context menu from contextMenu.js only disappears after user's click, ignoring non-user click events + * prevent Reload UI button/link from reloading the page when it's not yet ready + * fix prompts from file script failing to read contents from a drag/drop file + + +## 1.1.1 +### Bug Fixes: + * fix an error that prevents running webui on PyTorch<2.0 without --disable-safe-unpickle + +## 1.1.0 +### Features: + * switch to PyTorch 2.0.0 (except for AMD GPUs) + * visual improvements to custom code scripts + * add filename patterns: `[clip_skip]`, `[hasprompt<>]`, `[batch_number]`, `[generation_number]` + * add support for saving init images in img2img, and record their hashes in infotext for reproducability + * automatically select current word when adjusting weight with ctrl+up/down + * add dropdowns for X/Y/Z plot + * add setting: Stable Diffusion/Random number generator source: makes it possible to make images generated from a given manual seed consistent across different GPUs + * support Gradio's theme API + * use TCMalloc on Linux by default; possible fix for memory leaks + * add optimization option to remove negative conditioning at low sigma values #9177 + * embed model merge metadata in .safetensors file + * extension settings backup/restore feature #9169 + * add "resize by" and "resize to" tabs to img2img + * add option "keep original size" to textual inversion images preprocess + * image viewer scrolling via analog stick + * button to restore the progress from session lost / tab reload + +### Minor: + * bump Gradio to 3.28.1 + * change "scale to" to sliders in Extras tab + * add labels to tool buttons to make it possible to hide them + * add tiled inference support for ScuNET + * add branch support for extension installation + * change Linux installation script to install into current directory rather than `/home/username` + * sort textual inversion embeddings by name (case-insensitive) + * allow styles.csv to be symlinked or mounted in docker + * remove the "do not add watermark to images" option + * make selected tab configurable with UI config + * make the extra networks UI fixed height and scrollable + * add `disable_tls_verify` arg for use with self-signed certs + +### Extensions: + * add reload callback + * add `is_hr_pass` field for processing + +### Bug Fixes: + * fix broken batch image processing on 'Extras/Batch Process' tab + * add "None" option to extra networks dropdowns + * fix FileExistsError for CLIP Interrogator + * fix /sdapi/v1/txt2img endpoint not working on Linux #9319 + * fix disappearing live previews and progressbar during slow tasks + * fix fullscreen image view not working properly in some cases + * prevent alwayson_scripts args param resizing script_arg list when they are inserted in it + * fix prompt schedule for second order samplers + * fix image mask/composite for weird resolutions #9628 + * use correct images for previews when using AND (see #9491) + * one broken image in img2img batch won't stop all processing + * fix image orientation bug in train/preprocess + * fix Ngrok recreating tunnels every reload + * fix `--realesrgan-models-path` and `--ldsr-models-path` not working + * fix `--skip-install` not working + * use SAMPLE file format in Outpainting Mk2 & Poorman + * do not fail all LoRAs if some have failed to load when making a picture + +## 1.0.0 + * everything diff --git a/CODEOWNERS b/CODEOWNERS new file mode 100644 index 0000000000000000000000000000000000000000..2c937f6f1e519f864d15d5233e1fb86c6cdfac2f --- /dev/null +++ b/CODEOWNERS @@ -0,0 +1,12 @@ +* @AUTOMATIC1111 + +# if you were managing a localization and were removed from this file, this is because +# the intended way to do localizations now is via extensions. See: +# https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Developing-extensions +# Make a repo with your localization and since you are still listed as a collaborator +# you can add it to the wiki page yourself. This change is because some people complained +# the git commit log is cluttered with things unrelated to almost everyone and +# because I believe this is the best overall for the project to handle localizations almost +# entirely without my oversight. + + diff --git a/LICENSE.txt b/LICENSE.txt new file mode 100644 index 0000000000000000000000000000000000000000..211d32e752cb61bd056436e8f7a806f12a626bb7 --- /dev/null +++ b/LICENSE.txt @@ -0,0 +1,663 @@ + GNU AFFERO GENERAL PUBLIC LICENSE + Version 3, 19 November 2007 + + Copyright (c) 2023 AUTOMATIC1111 + + Copyright (C) 2007 Free Software Foundation, Inc. + Everyone is permitted to copy and distribute verbatim copies + of this license document, but changing it is not allowed. + + Preamble + + The GNU Affero General Public License is a free, copyleft license for +software and other kinds of works, specifically designed to ensure +cooperation with the community in the case of network server software. + + The licenses for most software and other practical works are designed +to take away your freedom to share and change the works. By contrast, +our General Public Licenses are intended to guarantee your freedom to +share and change all versions of a program--to make sure it remains free +software for all its users. + + When we speak of free software, we are referring to freedom, not +price. Our General Public Licenses are designed to make sure that you +have the freedom to distribute copies of free software (and charge for +them if you wish), that you receive source code or can get it if you +want it, that you can change the software or use pieces of it in new +free programs, and that you know you can do these things. + + Developers that use our General Public Licenses protect your rights +with two steps: (1) assert copyright on the software, and (2) offer +you this License which gives you legal permission to copy, distribute +and/or modify the software. + + A secondary benefit of defending all users' freedom is that +improvements made in alternate versions of the program, if they +receive widespread use, become available for other developers to +incorporate. Many developers of free software are heartened and +encouraged by the resulting cooperation. However, in the case of +software used on network servers, this result may fail to come about. +The GNU General Public License permits making a modified version and +letting the public access it on a server without ever releasing its +source code to the public. + + The GNU Affero General Public License is designed specifically to +ensure that, in such cases, the modified source code becomes available +to the community. It requires the operator of a network server to +provide the source code of the modified version running there to the +users of that server. Therefore, public use of a modified version, on +a publicly accessible server, gives the public access to the source +code of the modified version. + + An older license, called the Affero General Public License and +published by Affero, was designed to accomplish similar goals. This is +a different license, not a version of the Affero GPL, but Affero has +released a new version of the Affero GPL which permits relicensing under +this license. + + The precise terms and conditions for copying, distribution and +modification follow. + + TERMS AND CONDITIONS + + 0. Definitions. + + "This License" refers to version 3 of the GNU Affero General Public License. + + "Copyright" also means copyright-like laws that apply to other kinds of +works, such as semiconductor masks. + + "The Program" refers to any copyrightable work licensed under this +License. Each licensee is addressed as "you". "Licensees" and +"recipients" may be individuals or organizations. + + To "modify" a work means to copy from or adapt all or part of the work +in a fashion requiring copyright permission, other than the making of an +exact copy. The resulting work is called a "modified version" of the +earlier work or a work "based on" the earlier work. + + A "covered work" means either the unmodified Program or a work based +on the Program. + + To "propagate" a work means to do anything with it that, without +permission, would make you directly or secondarily liable for +infringement under applicable copyright law, except executing it on a +computer or modifying a private copy. Propagation includes copying, +distribution (with or without modification), making available to the +public, and in some countries other activities as well. + + To "convey" a work means any kind of propagation that enables other +parties to make or receive copies. Mere interaction with a user through +a computer network, with no transfer of a copy, is not conveying. + + An interactive user interface displays "Appropriate Legal Notices" +to the extent that it includes a convenient and prominently visible +feature that (1) displays an appropriate copyright notice, and (2) +tells the user that there is no warranty for the work (except to the +extent that warranties are provided), that licensees may convey the +work under this License, and how to view a copy of this License. If +the interface presents a list of user commands or options, such as a +menu, a prominent item in the list meets this criterion. + + 1. Source Code. + + The "source code" for a work means the preferred form of the work +for making modifications to it. "Object code" means any non-source +form of a work. + + A "Standard Interface" means an interface that either is an official +standard defined by a recognized standards body, or, in the case of +interfaces specified for a particular programming language, one that +is widely used among developers working in that language. + + The "System Libraries" of an executable work include anything, other +than the work as a whole, that (a) is included in the normal form of +packaging a Major Component, but which is not part of that Major +Component, and (b) serves only to enable use of the work with that +Major Component, or to implement a Standard Interface for which an +implementation is available to the public in source code form. A +"Major Component", in this context, means a major essential component +(kernel, window system, and so on) of the specific operating system +(if any) on which the executable work runs, or a compiler used to +produce the work, or an object code interpreter used to run it. + + The "Corresponding Source" for a work in object code form means all +the source code needed to generate, install, and (for an executable +work) run the object code and to modify the work, including scripts to +control those activities. However, it does not include the work's +System Libraries, or general-purpose tools or generally available free +programs which are used unmodified in performing those activities but +which are not part of the work. For example, Corresponding Source +includes interface definition files associated with source files for +the work, and the source code for shared libraries and dynamically +linked subprograms that the work is specifically designed to require, +such as by intimate data communication or control flow between those +subprograms and other parts of the work. + + The Corresponding Source need not include anything that users +can regenerate automatically from other parts of the Corresponding +Source. + + The Corresponding Source for a work in source code form is that +same work. + + 2. Basic Permissions. + + All rights granted under this License are granted for the term of +copyright on the Program, and are irrevocable provided the stated +conditions are met. This License explicitly affirms your unlimited +permission to run the unmodified Program. The output from running a +covered work is covered by this License only if the output, given its +content, constitutes a covered work. This License acknowledges your +rights of fair use or other equivalent, as provided by copyright law. + + You may make, run and propagate covered works that you do not +convey, without conditions so long as your license otherwise remains +in force. You may convey covered works to others for the sole purpose +of having them make modifications exclusively for you, or provide you +with facilities for running those works, provided that you comply with +the terms of this License in conveying all material for which you do +not control copyright. Those thus making or running the covered works +for you must do so exclusively on your behalf, under your direction +and control, on terms that prohibit them from making any copies of +your copyrighted material outside their relationship with you. + + Conveying under any other circumstances is permitted solely under +the conditions stated below. Sublicensing is not allowed; section 10 +makes it unnecessary. + + 3. Protecting Users' Legal Rights From Anti-Circumvention Law. + + No covered work shall be deemed part of an effective technological +measure under any applicable law fulfilling obligations under article +11 of the WIPO copyright treaty adopted on 20 December 1996, or +similar laws prohibiting or restricting circumvention of such +measures. + + When you convey a covered work, you waive any legal power to forbid +circumvention of technological measures to the extent such circumvention +is effected by exercising rights under this License with respect to +the covered work, and you disclaim any intention to limit operation or +modification of the work as a means of enforcing, against the work's +users, your or third parties' legal rights to forbid circumvention of +technological measures. + + 4. Conveying Verbatim Copies. + + You may convey verbatim copies of the Program's source code as you +receive it, in any medium, provided that you conspicuously and +appropriately publish on each copy an appropriate copyright notice; +keep intact all notices stating that this License and any +non-permissive terms added in accord with section 7 apply to the code; +keep intact all notices of the absence of any warranty; and give all +recipients a copy of this License along with the Program. + + You may charge any price or no price for each copy that you convey, +and you may offer support or warranty protection for a fee. + + 5. Conveying Modified Source Versions. + + You may convey a work based on the Program, or the modifications to +produce it from the Program, in the form of source code under the +terms of section 4, provided that you also meet all of these conditions: + + a) The work must carry prominent notices stating that you modified + it, and giving a relevant date. + + b) The work must carry prominent notices stating that it is + released under this License and any conditions added under section + 7. This requirement modifies the requirement in section 4 to + "keep intact all notices". + + c) You must license the entire work, as a whole, under this + License to anyone who comes into possession of a copy. This + License will therefore apply, along with any applicable section 7 + additional terms, to the whole of the work, and all its parts, + regardless of how they are packaged. This License gives no + permission to license the work in any other way, but it does not + invalidate such permission if you have separately received it. + + d) If the work has interactive user interfaces, each must display + Appropriate Legal Notices; however, if the Program has interactive + interfaces that do not display Appropriate Legal Notices, your + work need not make them do so. + + A compilation of a covered work with other separate and independent +works, which are not by their nature extensions of the covered work, +and which are not combined with it such as to form a larger program, +in or on a volume of a storage or distribution medium, is called an +"aggregate" if the compilation and its resulting copyright are not +used to limit the access or legal rights of the compilation's users +beyond what the individual works permit. Inclusion of a covered work +in an aggregate does not cause this License to apply to the other +parts of the aggregate. + + 6. Conveying Non-Source Forms. + + You may convey a covered work in object code form under the terms +of sections 4 and 5, provided that you also convey the +machine-readable Corresponding Source under the terms of this License, +in one of these ways: + + a) Convey the object code in, or embodied in, a physical product + (including a physical distribution medium), accompanied by the + Corresponding Source fixed on a durable physical medium + customarily used for software interchange. + + b) Convey the object code in, or embodied in, a physical product + (including a physical distribution medium), accompanied by a + written offer, valid for at least three years and valid for as + long as you offer spare parts or customer support for that product + model, to give anyone who possesses the object code either (1) a + copy of the Corresponding Source for all the software in the + product that is covered by this License, on a durable physical + medium customarily used for software interchange, for a price no + more than your reasonable cost of physically performing this + conveying of source, or (2) access to copy the + Corresponding Source from a network server at no charge. + + c) Convey individual copies of the object code with a copy of the + written offer to provide the Corresponding Source. This + alternative is allowed only occasionally and noncommercially, and + only if you received the object code with such an offer, in accord + with subsection 6b. + + d) Convey the object code by offering access from a designated + place (gratis or for a charge), and offer equivalent access to the + Corresponding Source in the same way through the same place at no + further charge. You need not require recipients to copy the + Corresponding Source along with the object code. If the place to + copy the object code is a network server, the Corresponding Source + may be on a different server (operated by you or a third party) + that supports equivalent copying facilities, provided you maintain + clear directions next to the object code saying where to find the + Corresponding Source. Regardless of what server hosts the + Corresponding Source, you remain obligated to ensure that it is + available for as long as needed to satisfy these requirements. + + e) Convey the object code using peer-to-peer transmission, provided + you inform other peers where the object code and Corresponding + Source of the work are being offered to the general public at no + charge under subsection 6d. + + A separable portion of the object code, whose source code is excluded +from the Corresponding Source as a System Library, need not be +included in conveying the object code work. + + A "User Product" is either (1) a "consumer product", which means any +tangible personal property which is normally used for personal, family, +or household purposes, or (2) anything designed or sold for incorporation +into a dwelling. In determining whether a product is a consumer product, +doubtful cases shall be resolved in favor of coverage. For a particular +product received by a particular user, "normally used" refers to a +typical or common use of that class of product, regardless of the status +of the particular user or of the way in which the particular user +actually uses, or expects or is expected to use, the product. A product +is a consumer product regardless of whether the product has substantial +commercial, industrial or non-consumer uses, unless such uses represent +the only significant mode of use of the product. + + "Installation Information" for a User Product means any methods, +procedures, authorization keys, or other information required to install +and execute modified versions of a covered work in that User Product from +a modified version of its Corresponding Source. The information must +suffice to ensure that the continued functioning of the modified object +code is in no case prevented or interfered with solely because +modification has been made. + + If you convey an object code work under this section in, or with, or +specifically for use in, a User Product, and the conveying occurs as +part of a transaction in which the right of possession and use of the +User Product is transferred to the recipient in perpetuity or for a +fixed term (regardless of how the transaction is characterized), the +Corresponding Source conveyed under this section must be accompanied +by the Installation Information. But this requirement does not apply +if neither you nor any third party retains the ability to install +modified object code on the User Product (for example, the work has +been installed in ROM). + + The requirement to provide Installation Information does not include a +requirement to continue to provide support service, warranty, or updates +for a work that has been modified or installed by the recipient, or for +the User Product in which it has been modified or installed. Access to a +network may be denied when the modification itself materially and +adversely affects the operation of the network or violates the rules and +protocols for communication across the network. + + Corresponding Source conveyed, and Installation Information provided, +in accord with this section must be in a format that is publicly +documented (and with an implementation available to the public in +source code form), and must require no special password or key for +unpacking, reading or copying. + + 7. Additional Terms. + + "Additional permissions" are terms that supplement the terms of this +License by making exceptions from one or more of its conditions. +Additional permissions that are applicable to the entire Program shall +be treated as though they were included in this License, to the extent +that they are valid under applicable law. If additional permissions +apply only to part of the Program, that part may be used separately +under those permissions, but the entire Program remains governed by +this License without regard to the additional permissions. + + When you convey a copy of a covered work, you may at your option +remove any additional permissions from that copy, or from any part of +it. (Additional permissions may be written to require their own +removal in certain cases when you modify the work.) You may place +additional permissions on material, added by you to a covered work, +for which you have or can give appropriate copyright permission. + + Notwithstanding any other provision of this License, for material you +add to a covered work, you may (if authorized by the copyright holders of +that material) supplement the terms of this License with terms: + + a) Disclaiming warranty or limiting liability differently from the + terms of sections 15 and 16 of this License; or + + b) Requiring preservation of specified reasonable legal notices or + author attributions in that material or in the Appropriate Legal + Notices displayed by works containing it; or + + c) Prohibiting misrepresentation of the origin of that material, or + requiring that modified versions of such material be marked in + reasonable ways as different from the original version; or + + d) Limiting the use for publicity purposes of names of licensors or + authors of the material; or + + e) Declining to grant rights under trademark law for use of some + trade names, trademarks, or service marks; or + + f) Requiring indemnification of licensors and authors of that + material by anyone who conveys the material (or modified versions of + it) with contractual assumptions of liability to the recipient, for + any liability that these contractual assumptions directly impose on + those licensors and authors. + + All other non-permissive additional terms are considered "further +restrictions" within the meaning of section 10. If the Program as you +received it, or any part of it, contains a notice stating that it is +governed by this License along with a term that is a further +restriction, you may remove that term. If a license document contains +a further restriction but permits relicensing or conveying under this +License, you may add to a covered work material governed by the terms +of that license document, provided that the further restriction does +not survive such relicensing or conveying. + + If you add terms to a covered work in accord with this section, you +must place, in the relevant source files, a statement of the +additional terms that apply to those files, or a notice indicating +where to find the applicable terms. + + Additional terms, permissive or non-permissive, may be stated in the +form of a separately written license, or stated as exceptions; +the above requirements apply either way. + + 8. Termination. + + You may not propagate or modify a covered work except as expressly +provided under this License. Any attempt otherwise to propagate or +modify it is void, and will automatically terminate your rights under +this License (including any patent licenses granted under the third +paragraph of section 11). + + However, if you cease all violation of this License, then your +license from a particular copyright holder is reinstated (a) +provisionally, unless and until the copyright holder explicitly and +finally terminates your license, and (b) permanently, if the copyright +holder fails to notify you of the violation by some reasonable means +prior to 60 days after the cessation. + + Moreover, your license from a particular copyright holder is +reinstated permanently if the copyright holder notifies you of the +violation by some reasonable means, this is the first time you have +received notice of violation of this License (for any work) from that +copyright holder, and you cure the violation prior to 30 days after +your receipt of the notice. + + Termination of your rights under this section does not terminate the +licenses of parties who have received copies or rights from you under +this License. If your rights have been terminated and not permanently +reinstated, you do not qualify to receive new licenses for the same +material under section 10. + + 9. Acceptance Not Required for Having Copies. + + You are not required to accept this License in order to receive or +run a copy of the Program. Ancillary propagation of a covered work +occurring solely as a consequence of using peer-to-peer transmission +to receive a copy likewise does not require acceptance. However, +nothing other than this License grants you permission to propagate or +modify any covered work. These actions infringe copyright if you do +not accept this License. Therefore, by modifying or propagating a +covered work, you indicate your acceptance of this License to do so. + + 10. Automatic Licensing of Downstream Recipients. + + Each time you convey a covered work, the recipient automatically +receives a license from the original licensors, to run, modify and +propagate that work, subject to this License. You are not responsible +for enforcing compliance by third parties with this License. + + An "entity transaction" is a transaction transferring control of an +organization, or substantially all assets of one, or subdividing an +organization, or merging organizations. If propagation of a covered +work results from an entity transaction, each party to that +transaction who receives a copy of the work also receives whatever +licenses to the work the party's predecessor in interest had or could +give under the previous paragraph, plus a right to possession of the +Corresponding Source of the work from the predecessor in interest, if +the predecessor has it or can get it with reasonable efforts. + + You may not impose any further restrictions on the exercise of the +rights granted or affirmed under this License. For example, you may +not impose a license fee, royalty, or other charge for exercise of +rights granted under this License, and you may not initiate litigation +(including a cross-claim or counterclaim in a lawsuit) alleging that +any patent claim is infringed by making, using, selling, offering for +sale, or importing the Program or any portion of it. + + 11. Patents. + + A "contributor" is a copyright holder who authorizes use under this +License of the Program or a work on which the Program is based. The +work thus licensed is called the contributor's "contributor version". + + A contributor's "essential patent claims" are all patent claims +owned or controlled by the contributor, whether already acquired or +hereafter acquired, that would be infringed by some manner, permitted +by this License, of making, using, or selling its contributor version, +but do not include claims that would be infringed only as a +consequence of further modification of the contributor version. For +purposes of this definition, "control" includes the right to grant +patent sublicenses in a manner consistent with the requirements of +this License. + + Each contributor grants you a non-exclusive, worldwide, royalty-free +patent license under the contributor's essential patent claims, to +make, use, sell, offer for sale, import and otherwise run, modify and +propagate the contents of its contributor version. + + In the following three paragraphs, a "patent license" is any express +agreement or commitment, however denominated, not to enforce a patent +(such as an express permission to practice a patent or covenant not to +sue for patent infringement). To "grant" such a patent license to a +party means to make such an agreement or commitment not to enforce a +patent against the party. + + If you convey a covered work, knowingly relying on a patent license, +and the Corresponding Source of the work is not available for anyone +to copy, free of charge and under the terms of this License, through a +publicly available network server or other readily accessible means, +then you must either (1) cause the Corresponding Source to be so +available, or (2) arrange to deprive yourself of the benefit of the +patent license for this particular work, or (3) arrange, in a manner +consistent with the requirements of this License, to extend the patent +license to downstream recipients. "Knowingly relying" means you have +actual knowledge that, but for the patent license, your conveying the +covered work in a country, or your recipient's use of the covered work +in a country, would infringe one or more identifiable patents in that +country that you have reason to believe are valid. + + If, pursuant to or in connection with a single transaction or +arrangement, you convey, or propagate by procuring conveyance of, a +covered work, and grant a patent license to some of the parties +receiving the covered work authorizing them to use, propagate, modify +or convey a specific copy of the covered work, then the patent license +you grant is automatically extended to all recipients of the covered +work and works based on it. + + A patent license is "discriminatory" if it does not include within +the scope of its coverage, prohibits the exercise of, or is +conditioned on the non-exercise of one or more of the rights that are +specifically granted under this License. You may not convey a covered +work if you are a party to an arrangement with a third party that is +in the business of distributing software, under which you make payment +to the third party based on the extent of your activity of conveying +the work, and under which the third party grants, to any of the +parties who would receive the covered work from you, a discriminatory +patent license (a) in connection with copies of the covered work +conveyed by you (or copies made from those copies), or (b) primarily +for and in connection with specific products or compilations that +contain the covered work, unless you entered into that arrangement, +or that patent license was granted, prior to 28 March 2007. + + Nothing in this License shall be construed as excluding or limiting +any implied license or other defenses to infringement that may +otherwise be available to you under applicable patent law. + + 12. No Surrender of Others' Freedom. + + If conditions are imposed on you (whether by court order, agreement or +otherwise) that contradict the conditions of this License, they do not +excuse you from the conditions of this License. If you cannot convey a +covered work so as to satisfy simultaneously your obligations under this +License and any other pertinent obligations, then as a consequence you may +not convey it at all. For example, if you agree to terms that obligate you +to collect a royalty for further conveying from those to whom you convey +the Program, the only way you could satisfy both those terms and this +License would be to refrain entirely from conveying the Program. + + 13. Remote Network Interaction; Use with the GNU General Public License. + + Notwithstanding any other provision of this License, if you modify the +Program, your modified version must prominently offer all users +interacting with it remotely through a computer network (if your version +supports such interaction) an opportunity to receive the Corresponding +Source of your version by providing access to the Corresponding Source +from a network server at no charge, through some standard or customary +means of facilitating copying of software. This Corresponding Source +shall include the Corresponding Source for any work covered by version 3 +of the GNU General Public License that is incorporated pursuant to the +following paragraph. + + Notwithstanding any other provision of this License, you have +permission to link or combine any covered work with a work licensed +under version 3 of the GNU General Public License into a single +combined work, and to convey the resulting work. The terms of this +License will continue to apply to the part which is the covered work, +but the work with which it is combined will remain governed by version +3 of the GNU General Public License. + + 14. Revised Versions of this License. + + The Free Software Foundation may publish revised and/or new versions of +the GNU Affero General Public License from time to time. Such new versions +will be similar in spirit to the present version, but may differ in detail to +address new problems or concerns. + + Each version is given a distinguishing version number. If the +Program specifies that a certain numbered version of the GNU Affero General +Public License "or any later version" applies to it, you have the +option of following the terms and conditions either of that numbered +version or of any later version published by the Free Software +Foundation. If the Program does not specify a version number of the +GNU Affero General Public License, you may choose any version ever published +by the Free Software Foundation. + + If the Program specifies that a proxy can decide which future +versions of the GNU Affero General Public License can be used, that proxy's +public statement of acceptance of a version permanently authorizes you +to choose that version for the Program. + + Later license versions may give you additional or different +permissions. However, no additional obligations are imposed on any +author or copyright holder as a result of your choosing to follow a +later version. + + 15. Disclaimer of Warranty. + + THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY +APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT +HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY +OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, +THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR +PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM +IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF +ALL NECESSARY SERVICING, REPAIR OR CORRECTION. + + 16. Limitation of Liability. + + IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING +WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS +THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY +GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE +USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF +DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD +PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), +EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF +SUCH DAMAGES. + + 17. Interpretation of Sections 15 and 16. + + If the disclaimer of warranty and limitation of liability provided +above cannot be given local legal effect according to their terms, +reviewing courts shall apply local law that most closely approximates +an absolute waiver of all civil liability in connection with the +Program, unless a warranty or assumption of liability accompanies a +copy of the Program in return for a fee. + + END OF TERMS AND CONDITIONS + + How to Apply These Terms to Your New Programs + + If you develop a new program, and you want it to be of the greatest +possible use to the public, the best way to achieve this is to make it +free software which everyone can redistribute and change under these terms. + + To do so, attach the following notices to the program. It is safest +to attach them to the start of each source file to most effectively +state the exclusion of warranty; and each file should have at least +the "copyright" line and a pointer to where the full notice is found. + + + Copyright (C) + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU Affero General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU Affero General Public License for more details. + + You should have received a copy of the GNU Affero General Public License + along with this program. If not, see . + +Also add information on how to contact you by electronic and paper mail. + + If your software can interact with users remotely through a computer +network, you should also make sure that it provides a way for users to +get its source. For example, if your program is a web application, its +interface could display a "Source" link that leads users to an archive +of the code. There are many ways you could offer source, and different +solutions will be better for different programs; see section 13 for the +specific requirements. + + You should also get your employer (if you work as a programmer) or school, +if any, to sign a "copyright disclaimer" for the program, if necessary. +For more information on this, and how to apply and follow the GNU AGPL, see +. diff --git a/README.md b/README.md index 7689c723b7a20d33fc7ec51e0ac55c95037f6011..57d282ffa736899a8e49235d40fcec24c33a5c2c 100644 --- a/README.md +++ b/README.md @@ -4,7 +4,14 @@ emoji: 💻 colorFrom: red colorTo: blue sdk: docker -pinned: false +pinned: true +python_version: 3.10.6 +suggested_hardware: cpu-upgrade +suggested_storage: large +app_file : webui.bat +app_port : 7860 + + --- Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference diff --git a/configs/alt-diffusion-inference.yaml b/configs/alt-diffusion-inference.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cfbee72d71bfd7deed2075e423ca51bd1da0521c --- /dev/null +++ b/configs/alt-diffusion-inference.yaml @@ -0,0 +1,72 @@ +model: + base_learning_rate: 1.0e-04 + target: ldm.models.diffusion.ddpm.LatentDiffusion + params: + linear_start: 0.00085 + linear_end: 0.0120 + num_timesteps_cond: 1 + log_every_t: 200 + timesteps: 1000 + first_stage_key: "jpg" + cond_stage_key: "txt" + image_size: 64 + channels: 4 + cond_stage_trainable: false # Note: different from the one we trained before + conditioning_key: crossattn + monitor: val/loss_simple_ema + scale_factor: 0.18215 + use_ema: False + + scheduler_config: # 10000 warmup steps + target: ldm.lr_scheduler.LambdaLinearScheduler + params: + warm_up_steps: [ 10000 ] + cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases + f_start: [ 1.e-6 ] + f_max: [ 1. ] + f_min: [ 1. ] + + unet_config: + target: ldm.modules.diffusionmodules.openaimodel.UNetModel + params: + image_size: 32 # unused + in_channels: 4 + out_channels: 4 + model_channels: 320 + attention_resolutions: [ 4, 2, 1 ] + num_res_blocks: 2 + channel_mult: [ 1, 2, 4, 4 ] + num_heads: 8 + use_spatial_transformer: True + transformer_depth: 1 + context_dim: 768 + use_checkpoint: True + legacy: False + + first_stage_config: + target: ldm.models.autoencoder.AutoencoderKL + params: + embed_dim: 4 + monitor: val/rec_loss + ddconfig: + double_z: true + z_channels: 4 + resolution: 256 + in_channels: 3 + out_ch: 3 + ch: 128 + ch_mult: + - 1 + - 2 + - 4 + - 4 + num_res_blocks: 2 + attn_resolutions: [] + dropout: 0.0 + lossconfig: + target: torch.nn.Identity + + cond_stage_config: + target: modules.xlmr.BertSeriesModelWithTransformation + params: + name: "XLMR-Large" \ No newline at end of file diff --git a/configs/instruct-pix2pix.yaml b/configs/instruct-pix2pix.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4e896879dd7ac5697b89cb323ec43eb41c03596c --- /dev/null +++ b/configs/instruct-pix2pix.yaml @@ -0,0 +1,98 @@ +# File modified by authors of InstructPix2Pix from original (https://github.com/CompVis/stable-diffusion). +# See more details in LICENSE. + +model: + base_learning_rate: 1.0e-04 + target: modules.models.diffusion.ddpm_edit.LatentDiffusion + params: + linear_start: 0.00085 + linear_end: 0.0120 + num_timesteps_cond: 1 + log_every_t: 200 + timesteps: 1000 + first_stage_key: edited + cond_stage_key: edit + # image_size: 64 + # image_size: 32 + image_size: 16 + channels: 4 + cond_stage_trainable: false # Note: different from the one we trained before + conditioning_key: hybrid + monitor: val/loss_simple_ema + scale_factor: 0.18215 + use_ema: false + + scheduler_config: # 10000 warmup steps + target: ldm.lr_scheduler.LambdaLinearScheduler + params: + warm_up_steps: [ 0 ] + cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases + f_start: [ 1.e-6 ] + f_max: [ 1. ] + f_min: [ 1. ] + + unet_config: + target: ldm.modules.diffusionmodules.openaimodel.UNetModel + params: + image_size: 32 # unused + in_channels: 8 + out_channels: 4 + model_channels: 320 + attention_resolutions: [ 4, 2, 1 ] + num_res_blocks: 2 + channel_mult: [ 1, 2, 4, 4 ] + num_heads: 8 + use_spatial_transformer: True + transformer_depth: 1 + context_dim: 768 + use_checkpoint: True + legacy: False + + first_stage_config: + target: ldm.models.autoencoder.AutoencoderKL + params: + embed_dim: 4 + monitor: val/rec_loss + ddconfig: + double_z: true + z_channels: 4 + resolution: 256 + in_channels: 3 + out_ch: 3 + ch: 128 + ch_mult: + - 1 + - 2 + - 4 + - 4 + num_res_blocks: 2 + attn_resolutions: [] + dropout: 0.0 + lossconfig: + target: torch.nn.Identity + + cond_stage_config: + target: ldm.modules.encoders.modules.FrozenCLIPEmbedder + +data: + target: main.DataModuleFromConfig + params: + batch_size: 128 + num_workers: 1 + wrap: false + validation: + target: edit_dataset.EditDataset + params: + path: data/clip-filtered-dataset + cache_dir: data/ + cache_name: data_10k + split: val + min_text_sim: 0.2 + min_image_sim: 0.75 + min_direction_sim: 0.2 + max_samples_per_prompt: 1 + min_resize_res: 512 + max_resize_res: 512 + crop_res: 512 + output_as_edit: False + real_input: True diff --git a/configs/v1-inference.yaml b/configs/v1-inference.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d4effe569e897369918625f9d8be5603a0e6a0d6 --- /dev/null +++ b/configs/v1-inference.yaml @@ -0,0 +1,70 @@ +model: + base_learning_rate: 1.0e-04 + target: ldm.models.diffusion.ddpm.LatentDiffusion + params: + linear_start: 0.00085 + linear_end: 0.0120 + num_timesteps_cond: 1 + log_every_t: 200 + timesteps: 1000 + first_stage_key: "jpg" + cond_stage_key: "txt" + image_size: 64 + channels: 4 + cond_stage_trainable: false # Note: different from the one we trained before + conditioning_key: crossattn + monitor: val/loss_simple_ema + scale_factor: 0.18215 + use_ema: False + + scheduler_config: # 10000 warmup steps + target: ldm.lr_scheduler.LambdaLinearScheduler + params: + warm_up_steps: [ 10000 ] + cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases + f_start: [ 1.e-6 ] + f_max: [ 1. ] + f_min: [ 1. ] + + unet_config: + target: ldm.modules.diffusionmodules.openaimodel.UNetModel + params: + image_size: 32 # unused + in_channels: 4 + out_channels: 4 + model_channels: 320 + attention_resolutions: [ 4, 2, 1 ] + num_res_blocks: 2 + channel_mult: [ 1, 2, 4, 4 ] + num_heads: 8 + use_spatial_transformer: True + transformer_depth: 1 + context_dim: 768 + use_checkpoint: True + legacy: False + + first_stage_config: + target: ldm.models.autoencoder.AutoencoderKL + params: + embed_dim: 4 + monitor: val/rec_loss + ddconfig: + double_z: true + z_channels: 4 + resolution: 256 + in_channels: 3 + out_ch: 3 + ch: 128 + ch_mult: + - 1 + - 2 + - 4 + - 4 + num_res_blocks: 2 + attn_resolutions: [] + dropout: 0.0 + lossconfig: + target: torch.nn.Identity + + cond_stage_config: + target: ldm.modules.encoders.modules.FrozenCLIPEmbedder diff --git a/configs/v1-inpainting-inference.yaml b/configs/v1-inpainting-inference.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f9eec37d24bce33ce92320a782d16ae72308190a --- /dev/null +++ b/configs/v1-inpainting-inference.yaml @@ -0,0 +1,70 @@ +model: + base_learning_rate: 7.5e-05 + target: ldm.models.diffusion.ddpm.LatentInpaintDiffusion + params: + linear_start: 0.00085 + linear_end: 0.0120 + num_timesteps_cond: 1 + log_every_t: 200 + timesteps: 1000 + first_stage_key: "jpg" + cond_stage_key: "txt" + image_size: 64 + channels: 4 + cond_stage_trainable: false # Note: different from the one we trained before + conditioning_key: hybrid # important + monitor: val/loss_simple_ema + scale_factor: 0.18215 + finetune_keys: null + + scheduler_config: # 10000 warmup steps + target: ldm.lr_scheduler.LambdaLinearScheduler + params: + warm_up_steps: [ 2500 ] # NOTE for resuming. use 10000 if starting from scratch + cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases + f_start: [ 1.e-6 ] + f_max: [ 1. ] + f_min: [ 1. ] + + unet_config: + target: ldm.modules.diffusionmodules.openaimodel.UNetModel + params: + image_size: 32 # unused + in_channels: 9 # 4 data + 4 downscaled image + 1 mask + out_channels: 4 + model_channels: 320 + attention_resolutions: [ 4, 2, 1 ] + num_res_blocks: 2 + channel_mult: [ 1, 2, 4, 4 ] + num_heads: 8 + use_spatial_transformer: True + transformer_depth: 1 + context_dim: 768 + use_checkpoint: True + legacy: False + + first_stage_config: + target: ldm.models.autoencoder.AutoencoderKL + params: + embed_dim: 4 + monitor: val/rec_loss + ddconfig: + double_z: true + z_channels: 4 + resolution: 256 + in_channels: 3 + out_ch: 3 + ch: 128 + ch_mult: + - 1 + - 2 + - 4 + - 4 + num_res_blocks: 2 + attn_resolutions: [] + dropout: 0.0 + lossconfig: + target: torch.nn.Identity + + cond_stage_config: + target: ldm.modules.encoders.modules.FrozenCLIPEmbedder diff --git a/embeddings/Place Textual Inversion embeddings here.txt b/embeddings/Place Textual Inversion embeddings here.txt new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/environment-wsl2.yaml b/environment-wsl2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0c4ae6809997ec38e7cf62cf0f71360b8cb61a7e --- /dev/null +++ b/environment-wsl2.yaml @@ -0,0 +1,11 @@ +name: automatic +channels: + - pytorch + - defaults +dependencies: + - python=3.10 + - pip=23.0 + - cudatoolkit=11.8 + - pytorch=2.0 + - torchvision=0.15 + - numpy=1.23 diff --git a/extensions-builtin/LDSR/ldsr_model_arch.py b/extensions-builtin/LDSR/ldsr_model_arch.py new file mode 100644 index 0000000000000000000000000000000000000000..7f450086ff95e40d7d5cab630b9942f330a09579 --- /dev/null +++ b/extensions-builtin/LDSR/ldsr_model_arch.py @@ -0,0 +1,252 @@ +import os +import gc +import time + +import numpy as np +import torch +import torchvision +from PIL import Image +from einops import rearrange, repeat +from omegaconf import OmegaConf +import safetensors.torch + +from ldm.models.diffusion.ddim import DDIMSampler +from ldm.util import instantiate_from_config, ismap +from modules import shared, sd_hijack + +cached_ldsr_model: torch.nn.Module = None + + +# Create LDSR Class +class LDSR: + def load_model_from_config(self, half_attention): + global cached_ldsr_model + + if shared.opts.ldsr_cached and cached_ldsr_model is not None: + print("Loading model from cache") + model: torch.nn.Module = cached_ldsr_model + else: + print(f"Loading model from {self.modelPath}") + _, extension = os.path.splitext(self.modelPath) + if extension.lower() == ".safetensors": + pl_sd = safetensors.torch.load_file(self.modelPath, device="cpu") + else: + pl_sd = torch.load(self.modelPath, map_location="cpu") + sd = pl_sd["state_dict"] if "state_dict" in pl_sd else pl_sd + config = OmegaConf.load(self.yamlPath) + config.model.target = "ldm.models.diffusion.ddpm.LatentDiffusionV1" + model: torch.nn.Module = instantiate_from_config(config.model) + model.load_state_dict(sd, strict=False) + model = model.to(shared.device) + if half_attention: + model = model.half() + if shared.cmd_opts.opt_channelslast: + model = model.to(memory_format=torch.channels_last) + + sd_hijack.model_hijack.hijack(model) # apply optimization + model.eval() + + if shared.opts.ldsr_cached: + cached_ldsr_model = model + + return {"model": model} + + def __init__(self, model_path, yaml_path): + self.modelPath = model_path + self.yamlPath = yaml_path + + @staticmethod + def run(model, selected_path, custom_steps, eta): + example = get_cond(selected_path) + + n_runs = 1 + guider = None + ckwargs = None + ddim_use_x0_pred = False + temperature = 1. + eta = eta + custom_shape = None + + height, width = example["image"].shape[1:3] + split_input = height >= 128 and width >= 128 + + if split_input: + ks = 128 + stride = 64 + vqf = 4 # + model.split_input_params = {"ks": (ks, ks), "stride": (stride, stride), + "vqf": vqf, + "patch_distributed_vq": True, + "tie_braker": False, + "clip_max_weight": 0.5, + "clip_min_weight": 0.01, + "clip_max_tie_weight": 0.5, + "clip_min_tie_weight": 0.01} + else: + if hasattr(model, "split_input_params"): + delattr(model, "split_input_params") + + x_t = None + logs = None + for _ in range(n_runs): + if custom_shape is not None: + x_t = torch.randn(1, custom_shape[1], custom_shape[2], custom_shape[3]).to(model.device) + x_t = repeat(x_t, '1 c h w -> b c h w', b=custom_shape[0]) + + logs = make_convolutional_sample(example, model, + custom_steps=custom_steps, + eta=eta, quantize_x0=False, + custom_shape=custom_shape, + temperature=temperature, noise_dropout=0., + corrector=guider, corrector_kwargs=ckwargs, x_T=x_t, + ddim_use_x0_pred=ddim_use_x0_pred + ) + return logs + + def super_resolution(self, image, steps=100, target_scale=2, half_attention=False): + model = self.load_model_from_config(half_attention) + + # Run settings + diffusion_steps = int(steps) + eta = 1.0 + + + gc.collect() + if torch.cuda.is_available: + torch.cuda.empty_cache() + + im_og = image + width_og, height_og = im_og.size + # If we can adjust the max upscale size, then the 4 below should be our variable + down_sample_rate = target_scale / 4 + wd = width_og * down_sample_rate + hd = height_og * down_sample_rate + width_downsampled_pre = int(np.ceil(wd)) + height_downsampled_pre = int(np.ceil(hd)) + + if down_sample_rate != 1: + print( + f'Downsampling from [{width_og}, {height_og}] to [{width_downsampled_pre}, {height_downsampled_pre}]') + im_og = im_og.resize((width_downsampled_pre, height_downsampled_pre), Image.LANCZOS) + else: + print(f"Down sample rate is 1 from {target_scale} / 4 (Not downsampling)") + + # pad width and height to multiples of 64, pads with the edge values of image to avoid artifacts + pad_w, pad_h = np.max(((2, 2), np.ceil(np.array(im_og.size) / 64).astype(int)), axis=0) * 64 - im_og.size + im_padded = Image.fromarray(np.pad(np.array(im_og), ((0, pad_h), (0, pad_w), (0, 0)), mode='edge')) + + logs = self.run(model["model"], im_padded, diffusion_steps, eta) + + sample = logs["sample"] + sample = sample.detach().cpu() + sample = torch.clamp(sample, -1., 1.) + sample = (sample + 1.) / 2. * 255 + sample = sample.numpy().astype(np.uint8) + sample = np.transpose(sample, (0, 2, 3, 1)) + a = Image.fromarray(sample[0]) + + # remove padding + a = a.crop((0, 0) + tuple(np.array(im_og.size) * 4)) + + del model + gc.collect() + if torch.cuda.is_available: + torch.cuda.empty_cache() + + return a + + +def get_cond(selected_path): + example = {} + up_f = 4 + c = selected_path.convert('RGB') + c = torch.unsqueeze(torchvision.transforms.ToTensor()(c), 0) + c_up = torchvision.transforms.functional.resize(c, size=[up_f * c.shape[2], up_f * c.shape[3]], + antialias=True) + c_up = rearrange(c_up, '1 c h w -> 1 h w c') + c = rearrange(c, '1 c h w -> 1 h w c') + c = 2. * c - 1. + + c = c.to(shared.device) + example["LR_image"] = c + example["image"] = c_up + + return example + + +@torch.no_grad() +def convsample_ddim(model, cond, steps, shape, eta=1.0, callback=None, normals_sequence=None, + mask=None, x0=None, quantize_x0=False, temperature=1., score_corrector=None, + corrector_kwargs=None, x_t=None + ): + ddim = DDIMSampler(model) + bs = shape[0] + shape = shape[1:] + print(f"Sampling with eta = {eta}; steps: {steps}") + samples, intermediates = ddim.sample(steps, batch_size=bs, shape=shape, conditioning=cond, callback=callback, + normals_sequence=normals_sequence, quantize_x0=quantize_x0, eta=eta, + mask=mask, x0=x0, temperature=temperature, verbose=False, + score_corrector=score_corrector, + corrector_kwargs=corrector_kwargs, x_t=x_t) + + return samples, intermediates + + +@torch.no_grad() +def make_convolutional_sample(batch, model, custom_steps=None, eta=1.0, quantize_x0=False, custom_shape=None, temperature=1., noise_dropout=0., corrector=None, + corrector_kwargs=None, x_T=None, ddim_use_x0_pred=False): + log = {} + + z, c, x, xrec, xc = model.get_input(batch, model.first_stage_key, + return_first_stage_outputs=True, + force_c_encode=not (hasattr(model, 'split_input_params') + and model.cond_stage_key == 'coordinates_bbox'), + return_original_cond=True) + + if custom_shape is not None: + z = torch.randn(custom_shape) + print(f"Generating {custom_shape[0]} samples of shape {custom_shape[1:]}") + + z0 = None + + log["input"] = x + log["reconstruction"] = xrec + + if ismap(xc): + log["original_conditioning"] = model.to_rgb(xc) + if hasattr(model, 'cond_stage_key'): + log[model.cond_stage_key] = model.to_rgb(xc) + + else: + log["original_conditioning"] = xc if xc is not None else torch.zeros_like(x) + if model.cond_stage_model: + log[model.cond_stage_key] = xc if xc is not None else torch.zeros_like(x) + if model.cond_stage_key == 'class_label': + log[model.cond_stage_key] = xc[model.cond_stage_key] + + with model.ema_scope("Plotting"): + t0 = time.time() + + sample, intermediates = convsample_ddim(model, c, steps=custom_steps, shape=z.shape, + eta=eta, + quantize_x0=quantize_x0, mask=None, x0=z0, + temperature=temperature, score_corrector=corrector, corrector_kwargs=corrector_kwargs, + x_t=x_T) + t1 = time.time() + + if ddim_use_x0_pred: + sample = intermediates['pred_x0'][-1] + + x_sample = model.decode_first_stage(sample) + + try: + x_sample_noquant = model.decode_first_stage(sample, force_not_quantize=True) + log["sample_noquant"] = x_sample_noquant + log["sample_diff"] = torch.abs(x_sample_noquant - x_sample) + except Exception: + pass + + log["sample"] = x_sample + log["time"] = t1 - t0 + + return log diff --git a/extensions-builtin/LDSR/preload.py b/extensions-builtin/LDSR/preload.py new file mode 100644 index 0000000000000000000000000000000000000000..cfd478d545ed12ef74e73fa40b6defe0156859da --- /dev/null +++ b/extensions-builtin/LDSR/preload.py @@ -0,0 +1,6 @@ +import os +from modules import paths + + +def preload(parser): + parser.add_argument("--ldsr-models-path", type=str, help="Path to directory with LDSR model file(s).", default=os.path.join(paths.models_path, 'LDSR')) diff --git a/extensions-builtin/LDSR/scripts/ldsr_model.py b/extensions-builtin/LDSR/scripts/ldsr_model.py new file mode 100644 index 0000000000000000000000000000000000000000..c4da79f31f83f4d520b43f9291d964036e8bd4d1 --- /dev/null +++ b/extensions-builtin/LDSR/scripts/ldsr_model.py @@ -0,0 +1,76 @@ +import os +import sys +import traceback + +from basicsr.utils.download_util import load_file_from_url + +from modules.upscaler import Upscaler, UpscalerData +from ldsr_model_arch import LDSR +from modules import shared, script_callbacks +import sd_hijack_autoencoder # noqa: F401 +import sd_hijack_ddpm_v1 # noqa: F401 + + +class UpscalerLDSR(Upscaler): + def __init__(self, user_path): + self.name = "LDSR" + self.user_path = user_path + self.model_url = "https://heibox.uni-heidelberg.de/f/578df07c8fc04ffbadf3/?dl=1" + self.yaml_url = "https://heibox.uni-heidelberg.de/f/31a76b13ea27482981b4/?dl=1" + super().__init__() + scaler_data = UpscalerData("LDSR", None, self) + self.scalers = [scaler_data] + + def load_model(self, path: str): + # Remove incorrect project.yaml file if too big + yaml_path = os.path.join(self.model_path, "project.yaml") + old_model_path = os.path.join(self.model_path, "model.pth") + new_model_path = os.path.join(self.model_path, "model.ckpt") + + local_model_paths = self.find_models(ext_filter=[".ckpt", ".safetensors"]) + local_ckpt_path = next(iter([local_model for local_model in local_model_paths if local_model.endswith("model.ckpt")]), None) + local_safetensors_path = next(iter([local_model for local_model in local_model_paths if local_model.endswith("model.safetensors")]), None) + local_yaml_path = next(iter([local_model for local_model in local_model_paths if local_model.endswith("project.yaml")]), None) + + if os.path.exists(yaml_path): + statinfo = os.stat(yaml_path) + if statinfo.st_size >= 10485760: + print("Removing invalid LDSR YAML file.") + os.remove(yaml_path) + + if os.path.exists(old_model_path): + print("Renaming model from model.pth to model.ckpt") + os.rename(old_model_path, new_model_path) + + if local_safetensors_path is not None and os.path.exists(local_safetensors_path): + model = local_safetensors_path + else: + model = local_ckpt_path if local_ckpt_path is not None else load_file_from_url(url=self.model_url, model_dir=self.model_download_path, file_name="model.ckpt", progress=True) + + yaml = local_yaml_path if local_yaml_path is not None else load_file_from_url(url=self.yaml_url, model_dir=self.model_download_path, file_name="project.yaml", progress=True) + + try: + return LDSR(model, yaml) + + except Exception: + print("Error importing LDSR:", file=sys.stderr) + print(traceback.format_exc(), file=sys.stderr) + return None + + def do_upscale(self, img, path): + ldsr = self.load_model(path) + if ldsr is None: + print("NO LDSR!") + return img + ddim_steps = shared.opts.ldsr_steps + return ldsr.super_resolution(img, ddim_steps, self.scale) + + +def on_ui_settings(): + import gradio as gr + + shared.opts.add_option("ldsr_steps", shared.OptionInfo(100, "LDSR processing steps. Lower = faster", gr.Slider, {"minimum": 1, "maximum": 200, "step": 1}, section=('upscaling', "Upscaling"))) + shared.opts.add_option("ldsr_cached", shared.OptionInfo(False, "Cache LDSR model in memory", gr.Checkbox, {"interactive": True}, section=('upscaling', "Upscaling"))) + + +script_callbacks.on_ui_settings(on_ui_settings) diff --git a/extensions-builtin/LDSR/sd_hijack_autoencoder.py b/extensions-builtin/LDSR/sd_hijack_autoencoder.py new file mode 100644 index 0000000000000000000000000000000000000000..81c5101b7d7f14f65c51ae91e97b63ee3debb086 --- /dev/null +++ b/extensions-builtin/LDSR/sd_hijack_autoencoder.py @@ -0,0 +1,292 @@ +# The content of this file comes from the ldm/models/autoencoder.py file of the compvis/stable-diffusion repo +# The VQModel & VQModelInterface were subsequently removed from ldm/models/autoencoder.py when we moved to the stability-ai/stablediffusion repo +# As the LDSR upscaler relies on VQModel & VQModelInterface, the hijack aims to put them back into the ldm.models.autoencoder +import numpy as np +import torch +import pytorch_lightning as pl +import torch.nn.functional as F +from contextlib import contextmanager + +from torch.optim.lr_scheduler import LambdaLR + +from ldm.modules.ema import LitEma +from taming.modules.vqvae.quantize import VectorQuantizer2 as VectorQuantizer +from ldm.modules.diffusionmodules.model import Encoder, Decoder +from ldm.util import instantiate_from_config + +import ldm.models.autoencoder +from packaging import version + +class VQModel(pl.LightningModule): + def __init__(self, + ddconfig, + lossconfig, + n_embed, + embed_dim, + ckpt_path=None, + ignore_keys=None, + image_key="image", + colorize_nlabels=None, + monitor=None, + batch_resize_range=None, + scheduler_config=None, + lr_g_factor=1.0, + remap=None, + sane_index_shape=False, # tell vector quantizer to return indices as bhw + use_ema=False + ): + super().__init__() + self.embed_dim = embed_dim + self.n_embed = n_embed + self.image_key = image_key + self.encoder = Encoder(**ddconfig) + self.decoder = Decoder(**ddconfig) + self.loss = instantiate_from_config(lossconfig) + self.quantize = VectorQuantizer(n_embed, embed_dim, beta=0.25, + remap=remap, + sane_index_shape=sane_index_shape) + self.quant_conv = torch.nn.Conv2d(ddconfig["z_channels"], embed_dim, 1) + self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1) + if colorize_nlabels is not None: + assert type(colorize_nlabels)==int + self.register_buffer("colorize", torch.randn(3, colorize_nlabels, 1, 1)) + if monitor is not None: + self.monitor = monitor + self.batch_resize_range = batch_resize_range + if self.batch_resize_range is not None: + print(f"{self.__class__.__name__}: Using per-batch resizing in range {batch_resize_range}.") + + self.use_ema = use_ema + if self.use_ema: + self.model_ema = LitEma(self) + print(f"Keeping EMAs of {len(list(self.model_ema.buffers()))}.") + + if ckpt_path is not None: + self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys or []) + self.scheduler_config = scheduler_config + self.lr_g_factor = lr_g_factor + + @contextmanager + def ema_scope(self, context=None): + if self.use_ema: + self.model_ema.store(self.parameters()) + self.model_ema.copy_to(self) + if context is not None: + print(f"{context}: Switched to EMA weights") + try: + yield None + finally: + if self.use_ema: + self.model_ema.restore(self.parameters()) + if context is not None: + print(f"{context}: Restored training weights") + + def init_from_ckpt(self, path, ignore_keys=None): + sd = torch.load(path, map_location="cpu")["state_dict"] + keys = list(sd.keys()) + for k in keys: + for ik in ignore_keys or []: + if k.startswith(ik): + print("Deleting key {} from state_dict.".format(k)) + del sd[k] + missing, unexpected = self.load_state_dict(sd, strict=False) + print(f"Restored from {path} with {len(missing)} missing and {len(unexpected)} unexpected keys") + if len(missing) > 0: + print(f"Missing Keys: {missing}") + print(f"Unexpected Keys: {unexpected}") + + def on_train_batch_end(self, *args, **kwargs): + if self.use_ema: + self.model_ema(self) + + def encode(self, x): + h = self.encoder(x) + h = self.quant_conv(h) + quant, emb_loss, info = self.quantize(h) + return quant, emb_loss, info + + def encode_to_prequant(self, x): + h = self.encoder(x) + h = self.quant_conv(h) + return h + + def decode(self, quant): + quant = self.post_quant_conv(quant) + dec = self.decoder(quant) + return dec + + def decode_code(self, code_b): + quant_b = self.quantize.embed_code(code_b) + dec = self.decode(quant_b) + return dec + + def forward(self, input, return_pred_indices=False): + quant, diff, (_,_,ind) = self.encode(input) + dec = self.decode(quant) + if return_pred_indices: + return dec, diff, ind + return dec, diff + + def get_input(self, batch, k): + x = batch[k] + if len(x.shape) == 3: + x = x[..., None] + x = x.permute(0, 3, 1, 2).to(memory_format=torch.contiguous_format).float() + if self.batch_resize_range is not None: + lower_size = self.batch_resize_range[0] + upper_size = self.batch_resize_range[1] + if self.global_step <= 4: + # do the first few batches with max size to avoid later oom + new_resize = upper_size + else: + new_resize = np.random.choice(np.arange(lower_size, upper_size+16, 16)) + if new_resize != x.shape[2]: + x = F.interpolate(x, size=new_resize, mode="bicubic") + x = x.detach() + return x + + def training_step(self, batch, batch_idx, optimizer_idx): + # https://github.com/pytorch/pytorch/issues/37142 + # try not to fool the heuristics + x = self.get_input(batch, self.image_key) + xrec, qloss, ind = self(x, return_pred_indices=True) + + if optimizer_idx == 0: + # autoencode + aeloss, log_dict_ae = self.loss(qloss, x, xrec, optimizer_idx, self.global_step, + last_layer=self.get_last_layer(), split="train", + predicted_indices=ind) + + self.log_dict(log_dict_ae, prog_bar=False, logger=True, on_step=True, on_epoch=True) + return aeloss + + if optimizer_idx == 1: + # discriminator + discloss, log_dict_disc = self.loss(qloss, x, xrec, optimizer_idx, self.global_step, + last_layer=self.get_last_layer(), split="train") + self.log_dict(log_dict_disc, prog_bar=False, logger=True, on_step=True, on_epoch=True) + return discloss + + def validation_step(self, batch, batch_idx): + log_dict = self._validation_step(batch, batch_idx) + with self.ema_scope(): + self._validation_step(batch, batch_idx, suffix="_ema") + return log_dict + + def _validation_step(self, batch, batch_idx, suffix=""): + x = self.get_input(batch, self.image_key) + xrec, qloss, ind = self(x, return_pred_indices=True) + aeloss, log_dict_ae = self.loss(qloss, x, xrec, 0, + self.global_step, + last_layer=self.get_last_layer(), + split="val"+suffix, + predicted_indices=ind + ) + + discloss, log_dict_disc = self.loss(qloss, x, xrec, 1, + self.global_step, + last_layer=self.get_last_layer(), + split="val"+suffix, + predicted_indices=ind + ) + rec_loss = log_dict_ae[f"val{suffix}/rec_loss"] + self.log(f"val{suffix}/rec_loss", rec_loss, + prog_bar=True, logger=True, on_step=False, on_epoch=True, sync_dist=True) + self.log(f"val{suffix}/aeloss", aeloss, + prog_bar=True, logger=True, on_step=False, on_epoch=True, sync_dist=True) + if version.parse(pl.__version__) >= version.parse('1.4.0'): + del log_dict_ae[f"val{suffix}/rec_loss"] + self.log_dict(log_dict_ae) + self.log_dict(log_dict_disc) + return self.log_dict + + def configure_optimizers(self): + lr_d = self.learning_rate + lr_g = self.lr_g_factor*self.learning_rate + print("lr_d", lr_d) + print("lr_g", lr_g) + opt_ae = torch.optim.Adam(list(self.encoder.parameters())+ + list(self.decoder.parameters())+ + list(self.quantize.parameters())+ + list(self.quant_conv.parameters())+ + list(self.post_quant_conv.parameters()), + lr=lr_g, betas=(0.5, 0.9)) + opt_disc = torch.optim.Adam(self.loss.discriminator.parameters(), + lr=lr_d, betas=(0.5, 0.9)) + + if self.scheduler_config is not None: + scheduler = instantiate_from_config(self.scheduler_config) + + print("Setting up LambdaLR scheduler...") + scheduler = [ + { + 'scheduler': LambdaLR(opt_ae, lr_lambda=scheduler.schedule), + 'interval': 'step', + 'frequency': 1 + }, + { + 'scheduler': LambdaLR(opt_disc, lr_lambda=scheduler.schedule), + 'interval': 'step', + 'frequency': 1 + }, + ] + return [opt_ae, opt_disc], scheduler + return [opt_ae, opt_disc], [] + + def get_last_layer(self): + return self.decoder.conv_out.weight + + def log_images(self, batch, only_inputs=False, plot_ema=False, **kwargs): + log = {} + x = self.get_input(batch, self.image_key) + x = x.to(self.device) + if only_inputs: + log["inputs"] = x + return log + xrec, _ = self(x) + if x.shape[1] > 3: + # colorize with random projection + assert xrec.shape[1] > 3 + x = self.to_rgb(x) + xrec = self.to_rgb(xrec) + log["inputs"] = x + log["reconstructions"] = xrec + if plot_ema: + with self.ema_scope(): + xrec_ema, _ = self(x) + if x.shape[1] > 3: + xrec_ema = self.to_rgb(xrec_ema) + log["reconstructions_ema"] = xrec_ema + return log + + def to_rgb(self, x): + assert self.image_key == "segmentation" + if not hasattr(self, "colorize"): + self.register_buffer("colorize", torch.randn(3, x.shape[1], 1, 1).to(x)) + x = F.conv2d(x, weight=self.colorize) + x = 2.*(x-x.min())/(x.max()-x.min()) - 1. + return x + + +class VQModelInterface(VQModel): + def __init__(self, embed_dim, *args, **kwargs): + super().__init__(*args, embed_dim=embed_dim, **kwargs) + self.embed_dim = embed_dim + + def encode(self, x): + h = self.encoder(x) + h = self.quant_conv(h) + return h + + def decode(self, h, force_not_quantize=False): + # also go through quantization layer + if not force_not_quantize: + quant, emb_loss, info = self.quantize(h) + else: + quant = h + quant = self.post_quant_conv(quant) + dec = self.decoder(quant) + return dec + +ldm.models.autoencoder.VQModel = VQModel +ldm.models.autoencoder.VQModelInterface = VQModelInterface diff --git a/extensions-builtin/LDSR/sd_hijack_ddpm_v1.py b/extensions-builtin/LDSR/sd_hijack_ddpm_v1.py new file mode 100644 index 0000000000000000000000000000000000000000..631a08ef0902246b3cc4e35e391db3a79af18346 --- /dev/null +++ b/extensions-builtin/LDSR/sd_hijack_ddpm_v1.py @@ -0,0 +1,1443 @@ +# This script is copied from the compvis/stable-diffusion repo (aka the SD V1 repo) +# Original filename: ldm/models/diffusion/ddpm.py +# The purpose to reinstate the old DDPM logic which works with VQ, whereas the V2 one doesn't +# Some models such as LDSR require VQ to work correctly +# The classes are suffixed with "V1" and added back to the "ldm.models.diffusion.ddpm" module + +import torch +import torch.nn as nn +import numpy as np +import pytorch_lightning as pl +from torch.optim.lr_scheduler import LambdaLR +from einops import rearrange, repeat +from contextlib import contextmanager +from functools import partial +from tqdm import tqdm +from torchvision.utils import make_grid +from pytorch_lightning.utilities.distributed import rank_zero_only + +from ldm.util import log_txt_as_img, exists, default, ismap, isimage, mean_flat, count_params, instantiate_from_config +from ldm.modules.ema import LitEma +from ldm.modules.distributions.distributions import normal_kl, DiagonalGaussianDistribution +from ldm.models.autoencoder import VQModelInterface, IdentityFirstStage, AutoencoderKL +from ldm.modules.diffusionmodules.util import make_beta_schedule, extract_into_tensor, noise_like +from ldm.models.diffusion.ddim import DDIMSampler + +import ldm.models.diffusion.ddpm + +__conditioning_keys__ = {'concat': 'c_concat', + 'crossattn': 'c_crossattn', + 'adm': 'y'} + + +def disabled_train(self, mode=True): + """Overwrite model.train with this function to make sure train/eval mode + does not change anymore.""" + return self + + +def uniform_on_device(r1, r2, shape, device): + return (r1 - r2) * torch.rand(*shape, device=device) + r2 + + +class DDPMV1(pl.LightningModule): + # classic DDPM with Gaussian diffusion, in image space + def __init__(self, + unet_config, + timesteps=1000, + beta_schedule="linear", + loss_type="l2", + ckpt_path=None, + ignore_keys=None, + load_only_unet=False, + monitor="val/loss", + use_ema=True, + first_stage_key="image", + image_size=256, + channels=3, + log_every_t=100, + clip_denoised=True, + linear_start=1e-4, + linear_end=2e-2, + cosine_s=8e-3, + given_betas=None, + original_elbo_weight=0., + v_posterior=0., # weight for choosing posterior variance as sigma = (1-v) * beta_tilde + v * beta + l_simple_weight=1., + conditioning_key=None, + parameterization="eps", # all assuming fixed variance schedules + scheduler_config=None, + use_positional_encodings=False, + learn_logvar=False, + logvar_init=0., + ): + super().__init__() + assert parameterization in ["eps", "x0"], 'currently only supporting "eps" and "x0"' + self.parameterization = parameterization + print(f"{self.__class__.__name__}: Running in {self.parameterization}-prediction mode") + self.cond_stage_model = None + self.clip_denoised = clip_denoised + self.log_every_t = log_every_t + self.first_stage_key = first_stage_key + self.image_size = image_size # try conv? + self.channels = channels + self.use_positional_encodings = use_positional_encodings + self.model = DiffusionWrapperV1(unet_config, conditioning_key) + count_params(self.model, verbose=True) + self.use_ema = use_ema + if self.use_ema: + self.model_ema = LitEma(self.model) + print(f"Keeping EMAs of {len(list(self.model_ema.buffers()))}.") + + self.use_scheduler = scheduler_config is not None + if self.use_scheduler: + self.scheduler_config = scheduler_config + + self.v_posterior = v_posterior + self.original_elbo_weight = original_elbo_weight + self.l_simple_weight = l_simple_weight + + if monitor is not None: + self.monitor = monitor + if ckpt_path is not None: + self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys or [], only_model=load_only_unet) + + self.register_schedule(given_betas=given_betas, beta_schedule=beta_schedule, timesteps=timesteps, + linear_start=linear_start, linear_end=linear_end, cosine_s=cosine_s) + + self.loss_type = loss_type + + self.learn_logvar = learn_logvar + self.logvar = torch.full(fill_value=logvar_init, size=(self.num_timesteps,)) + if self.learn_logvar: + self.logvar = nn.Parameter(self.logvar, requires_grad=True) + + + def register_schedule(self, given_betas=None, beta_schedule="linear", timesteps=1000, + linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3): + if exists(given_betas): + betas = given_betas + else: + betas = make_beta_schedule(beta_schedule, timesteps, linear_start=linear_start, linear_end=linear_end, + cosine_s=cosine_s) + alphas = 1. - betas + alphas_cumprod = np.cumprod(alphas, axis=0) + alphas_cumprod_prev = np.append(1., alphas_cumprod[:-1]) + + timesteps, = betas.shape + self.num_timesteps = int(timesteps) + self.linear_start = linear_start + self.linear_end = linear_end + assert alphas_cumprod.shape[0] == self.num_timesteps, 'alphas have to be defined for each timestep' + + to_torch = partial(torch.tensor, dtype=torch.float32) + + self.register_buffer('betas', to_torch(betas)) + self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod)) + self.register_buffer('alphas_cumprod_prev', to_torch(alphas_cumprod_prev)) + + # calculations for diffusion q(x_t | x_{t-1}) and others + self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod))) + self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod))) + self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod))) + self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod))) + self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod - 1))) + + # calculations for posterior q(x_{t-1} | x_t, x_0) + posterior_variance = (1 - self.v_posterior) * betas * (1. - alphas_cumprod_prev) / ( + 1. - alphas_cumprod) + self.v_posterior * betas + # above: equal to 1. / (1. / (1. - alpha_cumprod_tm1) + alpha_t / beta_t) + self.register_buffer('posterior_variance', to_torch(posterior_variance)) + # below: log calculation clipped because the posterior variance is 0 at the beginning of the diffusion chain + self.register_buffer('posterior_log_variance_clipped', to_torch(np.log(np.maximum(posterior_variance, 1e-20)))) + self.register_buffer('posterior_mean_coef1', to_torch( + betas * np.sqrt(alphas_cumprod_prev) / (1. - alphas_cumprod))) + self.register_buffer('posterior_mean_coef2', to_torch( + (1. - alphas_cumprod_prev) * np.sqrt(alphas) / (1. - alphas_cumprod))) + + if self.parameterization == "eps": + lvlb_weights = self.betas ** 2 / ( + 2 * self.posterior_variance * to_torch(alphas) * (1 - self.alphas_cumprod)) + elif self.parameterization == "x0": + lvlb_weights = 0.5 * np.sqrt(torch.Tensor(alphas_cumprod)) / (2. * 1 - torch.Tensor(alphas_cumprod)) + else: + raise NotImplementedError("mu not supported") + # TODO how to choose this term + lvlb_weights[0] = lvlb_weights[1] + self.register_buffer('lvlb_weights', lvlb_weights, persistent=False) + assert not torch.isnan(self.lvlb_weights).all() + + @contextmanager + def ema_scope(self, context=None): + if self.use_ema: + self.model_ema.store(self.model.parameters()) + self.model_ema.copy_to(self.model) + if context is not None: + print(f"{context}: Switched to EMA weights") + try: + yield None + finally: + if self.use_ema: + self.model_ema.restore(self.model.parameters()) + if context is not None: + print(f"{context}: Restored training weights") + + def init_from_ckpt(self, path, ignore_keys=None, only_model=False): + sd = torch.load(path, map_location="cpu") + if "state_dict" in list(sd.keys()): + sd = sd["state_dict"] + keys = list(sd.keys()) + for k in keys: + for ik in ignore_keys or []: + if k.startswith(ik): + print("Deleting key {} from state_dict.".format(k)) + del sd[k] + missing, unexpected = self.load_state_dict(sd, strict=False) if not only_model else self.model.load_state_dict( + sd, strict=False) + print(f"Restored from {path} with {len(missing)} missing and {len(unexpected)} unexpected keys") + if len(missing) > 0: + print(f"Missing Keys: {missing}") + if len(unexpected) > 0: + print(f"Unexpected Keys: {unexpected}") + + def q_mean_variance(self, x_start, t): + """ + Get the distribution q(x_t | x_0). + :param x_start: the [N x C x ...] tensor of noiseless inputs. + :param t: the number of diffusion steps (minus 1). Here, 0 means one step. + :return: A tuple (mean, variance, log_variance), all of x_start's shape. + """ + mean = (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start) + variance = extract_into_tensor(1.0 - self.alphas_cumprod, t, x_start.shape) + log_variance = extract_into_tensor(self.log_one_minus_alphas_cumprod, t, x_start.shape) + return mean, variance, log_variance + + def predict_start_from_noise(self, x_t, t, noise): + return ( + extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t - + extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * noise + ) + + def q_posterior(self, x_start, x_t, t): + posterior_mean = ( + extract_into_tensor(self.posterior_mean_coef1, t, x_t.shape) * x_start + + extract_into_tensor(self.posterior_mean_coef2, t, x_t.shape) * x_t + ) + posterior_variance = extract_into_tensor(self.posterior_variance, t, x_t.shape) + posterior_log_variance_clipped = extract_into_tensor(self.posterior_log_variance_clipped, t, x_t.shape) + return posterior_mean, posterior_variance, posterior_log_variance_clipped + + def p_mean_variance(self, x, t, clip_denoised: bool): + model_out = self.model(x, t) + if self.parameterization == "eps": + x_recon = self.predict_start_from_noise(x, t=t, noise=model_out) + elif self.parameterization == "x0": + x_recon = model_out + if clip_denoised: + x_recon.clamp_(-1., 1.) + + model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t) + return model_mean, posterior_variance, posterior_log_variance + + @torch.no_grad() + def p_sample(self, x, t, clip_denoised=True, repeat_noise=False): + b, *_, device = *x.shape, x.device + model_mean, _, model_log_variance = self.p_mean_variance(x=x, t=t, clip_denoised=clip_denoised) + noise = noise_like(x.shape, device, repeat_noise) + # no noise when t == 0 + nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1))) + return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise + + @torch.no_grad() + def p_sample_loop(self, shape, return_intermediates=False): + device = self.betas.device + b = shape[0] + img = torch.randn(shape, device=device) + intermediates = [img] + for i in tqdm(reversed(range(0, self.num_timesteps)), desc='Sampling t', total=self.num_timesteps): + img = self.p_sample(img, torch.full((b,), i, device=device, dtype=torch.long), + clip_denoised=self.clip_denoised) + if i % self.log_every_t == 0 or i == self.num_timesteps - 1: + intermediates.append(img) + if return_intermediates: + return img, intermediates + return img + + @torch.no_grad() + def sample(self, batch_size=16, return_intermediates=False): + image_size = self.image_size + channels = self.channels + return self.p_sample_loop((batch_size, channels, image_size, image_size), + return_intermediates=return_intermediates) + + def q_sample(self, x_start, t, noise=None): + noise = default(noise, lambda: torch.randn_like(x_start)) + return (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start + + extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise) + + def get_loss(self, pred, target, mean=True): + if self.loss_type == 'l1': + loss = (target - pred).abs() + if mean: + loss = loss.mean() + elif self.loss_type == 'l2': + if mean: + loss = torch.nn.functional.mse_loss(target, pred) + else: + loss = torch.nn.functional.mse_loss(target, pred, reduction='none') + else: + raise NotImplementedError("unknown loss type '{loss_type}'") + + return loss + + def p_losses(self, x_start, t, noise=None): + noise = default(noise, lambda: torch.randn_like(x_start)) + x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise) + model_out = self.model(x_noisy, t) + + loss_dict = {} + if self.parameterization == "eps": + target = noise + elif self.parameterization == "x0": + target = x_start + else: + raise NotImplementedError(f"Paramterization {self.parameterization} not yet supported") + + loss = self.get_loss(model_out, target, mean=False).mean(dim=[1, 2, 3]) + + log_prefix = 'train' if self.training else 'val' + + loss_dict.update({f'{log_prefix}/loss_simple': loss.mean()}) + loss_simple = loss.mean() * self.l_simple_weight + + loss_vlb = (self.lvlb_weights[t] * loss).mean() + loss_dict.update({f'{log_prefix}/loss_vlb': loss_vlb}) + + loss = loss_simple + self.original_elbo_weight * loss_vlb + + loss_dict.update({f'{log_prefix}/loss': loss}) + + return loss, loss_dict + + def forward(self, x, *args, **kwargs): + # b, c, h, w, device, img_size, = *x.shape, x.device, self.image_size + # assert h == img_size and w == img_size, f'height and width of image must be {img_size}' + t = torch.randint(0, self.num_timesteps, (x.shape[0],), device=self.device).long() + return self.p_losses(x, t, *args, **kwargs) + + def get_input(self, batch, k): + x = batch[k] + if len(x.shape) == 3: + x = x[..., None] + x = rearrange(x, 'b h w c -> b c h w') + x = x.to(memory_format=torch.contiguous_format).float() + return x + + def shared_step(self, batch): + x = self.get_input(batch, self.first_stage_key) + loss, loss_dict = self(x) + return loss, loss_dict + + def training_step(self, batch, batch_idx): + loss, loss_dict = self.shared_step(batch) + + self.log_dict(loss_dict, prog_bar=True, + logger=True, on_step=True, on_epoch=True) + + self.log("global_step", self.global_step, + prog_bar=True, logger=True, on_step=True, on_epoch=False) + + if self.use_scheduler: + lr = self.optimizers().param_groups[0]['lr'] + self.log('lr_abs', lr, prog_bar=True, logger=True, on_step=True, on_epoch=False) + + return loss + + @torch.no_grad() + def validation_step(self, batch, batch_idx): + _, loss_dict_no_ema = self.shared_step(batch) + with self.ema_scope(): + _, loss_dict_ema = self.shared_step(batch) + loss_dict_ema = {key + '_ema': loss_dict_ema[key] for key in loss_dict_ema} + self.log_dict(loss_dict_no_ema, prog_bar=False, logger=True, on_step=False, on_epoch=True) + self.log_dict(loss_dict_ema, prog_bar=False, logger=True, on_step=False, on_epoch=True) + + def on_train_batch_end(self, *args, **kwargs): + if self.use_ema: + self.model_ema(self.model) + + def _get_rows_from_list(self, samples): + n_imgs_per_row = len(samples) + denoise_grid = rearrange(samples, 'n b c h w -> b n c h w') + denoise_grid = rearrange(denoise_grid, 'b n c h w -> (b n) c h w') + denoise_grid = make_grid(denoise_grid, nrow=n_imgs_per_row) + return denoise_grid + + @torch.no_grad() + def log_images(self, batch, N=8, n_row=2, sample=True, return_keys=None, **kwargs): + log = {} + x = self.get_input(batch, self.first_stage_key) + N = min(x.shape[0], N) + n_row = min(x.shape[0], n_row) + x = x.to(self.device)[:N] + log["inputs"] = x + + # get diffusion row + diffusion_row = [] + x_start = x[:n_row] + + for t in range(self.num_timesteps): + if t % self.log_every_t == 0 or t == self.num_timesteps - 1: + t = repeat(torch.tensor([t]), '1 -> b', b=n_row) + t = t.to(self.device).long() + noise = torch.randn_like(x_start) + x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise) + diffusion_row.append(x_noisy) + + log["diffusion_row"] = self._get_rows_from_list(diffusion_row) + + if sample: + # get denoise row + with self.ema_scope("Plotting"): + samples, denoise_row = self.sample(batch_size=N, return_intermediates=True) + + log["samples"] = samples + log["denoise_row"] = self._get_rows_from_list(denoise_row) + + if return_keys: + if np.intersect1d(list(log.keys()), return_keys).shape[0] == 0: + return log + else: + return {key: log[key] for key in return_keys} + return log + + def configure_optimizers(self): + lr = self.learning_rate + params = list(self.model.parameters()) + if self.learn_logvar: + params = params + [self.logvar] + opt = torch.optim.AdamW(params, lr=lr) + return opt + + +class LatentDiffusionV1(DDPMV1): + """main class""" + def __init__(self, + first_stage_config, + cond_stage_config, + num_timesteps_cond=None, + cond_stage_key="image", + cond_stage_trainable=False, + concat_mode=True, + cond_stage_forward=None, + conditioning_key=None, + scale_factor=1.0, + scale_by_std=False, + *args, **kwargs): + self.num_timesteps_cond = default(num_timesteps_cond, 1) + self.scale_by_std = scale_by_std + assert self.num_timesteps_cond <= kwargs['timesteps'] + # for backwards compatibility after implementation of DiffusionWrapper + if conditioning_key is None: + conditioning_key = 'concat' if concat_mode else 'crossattn' + if cond_stage_config == '__is_unconditional__': + conditioning_key = None + ckpt_path = kwargs.pop("ckpt_path", None) + ignore_keys = kwargs.pop("ignore_keys", []) + super().__init__(*args, conditioning_key=conditioning_key, **kwargs) + self.concat_mode = concat_mode + self.cond_stage_trainable = cond_stage_trainable + self.cond_stage_key = cond_stage_key + try: + self.num_downs = len(first_stage_config.params.ddconfig.ch_mult) - 1 + except Exception: + self.num_downs = 0 + if not scale_by_std: + self.scale_factor = scale_factor + else: + self.register_buffer('scale_factor', torch.tensor(scale_factor)) + self.instantiate_first_stage(first_stage_config) + self.instantiate_cond_stage(cond_stage_config) + self.cond_stage_forward = cond_stage_forward + self.clip_denoised = False + self.bbox_tokenizer = None + + self.restarted_from_ckpt = False + if ckpt_path is not None: + self.init_from_ckpt(ckpt_path, ignore_keys) + self.restarted_from_ckpt = True + + def make_cond_schedule(self, ): + self.cond_ids = torch.full(size=(self.num_timesteps,), fill_value=self.num_timesteps - 1, dtype=torch.long) + ids = torch.round(torch.linspace(0, self.num_timesteps - 1, self.num_timesteps_cond)).long() + self.cond_ids[:self.num_timesteps_cond] = ids + + @rank_zero_only + @torch.no_grad() + def on_train_batch_start(self, batch, batch_idx, dataloader_idx): + # only for very first batch + if self.scale_by_std and self.current_epoch == 0 and self.global_step == 0 and batch_idx == 0 and not self.restarted_from_ckpt: + assert self.scale_factor == 1., 'rather not use custom rescaling and std-rescaling simultaneously' + # set rescale weight to 1./std of encodings + print("### USING STD-RESCALING ###") + x = super().get_input(batch, self.first_stage_key) + x = x.to(self.device) + encoder_posterior = self.encode_first_stage(x) + z = self.get_first_stage_encoding(encoder_posterior).detach() + del self.scale_factor + self.register_buffer('scale_factor', 1. / z.flatten().std()) + print(f"setting self.scale_factor to {self.scale_factor}") + print("### USING STD-RESCALING ###") + + def register_schedule(self, + given_betas=None, beta_schedule="linear", timesteps=1000, + linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3): + super().register_schedule(given_betas, beta_schedule, timesteps, linear_start, linear_end, cosine_s) + + self.shorten_cond_schedule = self.num_timesteps_cond > 1 + if self.shorten_cond_schedule: + self.make_cond_schedule() + + def instantiate_first_stage(self, config): + model = instantiate_from_config(config) + self.first_stage_model = model.eval() + self.first_stage_model.train = disabled_train + for param in self.first_stage_model.parameters(): + param.requires_grad = False + + def instantiate_cond_stage(self, config): + if not self.cond_stage_trainable: + if config == "__is_first_stage__": + print("Using first stage also as cond stage.") + self.cond_stage_model = self.first_stage_model + elif config == "__is_unconditional__": + print(f"Training {self.__class__.__name__} as an unconditional model.") + self.cond_stage_model = None + # self.be_unconditional = True + else: + model = instantiate_from_config(config) + self.cond_stage_model = model.eval() + self.cond_stage_model.train = disabled_train + for param in self.cond_stage_model.parameters(): + param.requires_grad = False + else: + assert config != '__is_first_stage__' + assert config != '__is_unconditional__' + model = instantiate_from_config(config) + self.cond_stage_model = model + + def _get_denoise_row_from_list(self, samples, desc='', force_no_decoder_quantization=False): + denoise_row = [] + for zd in tqdm(samples, desc=desc): + denoise_row.append(self.decode_first_stage(zd.to(self.device), + force_not_quantize=force_no_decoder_quantization)) + n_imgs_per_row = len(denoise_row) + denoise_row = torch.stack(denoise_row) # n_log_step, n_row, C, H, W + denoise_grid = rearrange(denoise_row, 'n b c h w -> b n c h w') + denoise_grid = rearrange(denoise_grid, 'b n c h w -> (b n) c h w') + denoise_grid = make_grid(denoise_grid, nrow=n_imgs_per_row) + return denoise_grid + + def get_first_stage_encoding(self, encoder_posterior): + if isinstance(encoder_posterior, DiagonalGaussianDistribution): + z = encoder_posterior.sample() + elif isinstance(encoder_posterior, torch.Tensor): + z = encoder_posterior + else: + raise NotImplementedError(f"encoder_posterior of type '{type(encoder_posterior)}' not yet implemented") + return self.scale_factor * z + + def get_learned_conditioning(self, c): + if self.cond_stage_forward is None: + if hasattr(self.cond_stage_model, 'encode') and callable(self.cond_stage_model.encode): + c = self.cond_stage_model.encode(c) + if isinstance(c, DiagonalGaussianDistribution): + c = c.mode() + else: + c = self.cond_stage_model(c) + else: + assert hasattr(self.cond_stage_model, self.cond_stage_forward) + c = getattr(self.cond_stage_model, self.cond_stage_forward)(c) + return c + + def meshgrid(self, h, w): + y = torch.arange(0, h).view(h, 1, 1).repeat(1, w, 1) + x = torch.arange(0, w).view(1, w, 1).repeat(h, 1, 1) + + arr = torch.cat([y, x], dim=-1) + return arr + + def delta_border(self, h, w): + """ + :param h: height + :param w: width + :return: normalized distance to image border, + wtith min distance = 0 at border and max dist = 0.5 at image center + """ + lower_right_corner = torch.tensor([h - 1, w - 1]).view(1, 1, 2) + arr = self.meshgrid(h, w) / lower_right_corner + dist_left_up = torch.min(arr, dim=-1, keepdims=True)[0] + dist_right_down = torch.min(1 - arr, dim=-1, keepdims=True)[0] + edge_dist = torch.min(torch.cat([dist_left_up, dist_right_down], dim=-1), dim=-1)[0] + return edge_dist + + def get_weighting(self, h, w, Ly, Lx, device): + weighting = self.delta_border(h, w) + weighting = torch.clip(weighting, self.split_input_params["clip_min_weight"], + self.split_input_params["clip_max_weight"], ) + weighting = weighting.view(1, h * w, 1).repeat(1, 1, Ly * Lx).to(device) + + if self.split_input_params["tie_braker"]: + L_weighting = self.delta_border(Ly, Lx) + L_weighting = torch.clip(L_weighting, + self.split_input_params["clip_min_tie_weight"], + self.split_input_params["clip_max_tie_weight"]) + + L_weighting = L_weighting.view(1, 1, Ly * Lx).to(device) + weighting = weighting * L_weighting + return weighting + + def get_fold_unfold(self, x, kernel_size, stride, uf=1, df=1): # todo load once not every time, shorten code + """ + :param x: img of size (bs, c, h, w) + :return: n img crops of size (n, bs, c, kernel_size[0], kernel_size[1]) + """ + bs, nc, h, w = x.shape + + # number of crops in image + Ly = (h - kernel_size[0]) // stride[0] + 1 + Lx = (w - kernel_size[1]) // stride[1] + 1 + + if uf == 1 and df == 1: + fold_params = dict(kernel_size=kernel_size, dilation=1, padding=0, stride=stride) + unfold = torch.nn.Unfold(**fold_params) + + fold = torch.nn.Fold(output_size=x.shape[2:], **fold_params) + + weighting = self.get_weighting(kernel_size[0], kernel_size[1], Ly, Lx, x.device).to(x.dtype) + normalization = fold(weighting).view(1, 1, h, w) # normalizes the overlap + weighting = weighting.view((1, 1, kernel_size[0], kernel_size[1], Ly * Lx)) + + elif uf > 1 and df == 1: + fold_params = dict(kernel_size=kernel_size, dilation=1, padding=0, stride=stride) + unfold = torch.nn.Unfold(**fold_params) + + fold_params2 = dict(kernel_size=(kernel_size[0] * uf, kernel_size[0] * uf), + dilation=1, padding=0, + stride=(stride[0] * uf, stride[1] * uf)) + fold = torch.nn.Fold(output_size=(x.shape[2] * uf, x.shape[3] * uf), **fold_params2) + + weighting = self.get_weighting(kernel_size[0] * uf, kernel_size[1] * uf, Ly, Lx, x.device).to(x.dtype) + normalization = fold(weighting).view(1, 1, h * uf, w * uf) # normalizes the overlap + weighting = weighting.view((1, 1, kernel_size[0] * uf, kernel_size[1] * uf, Ly * Lx)) + + elif df > 1 and uf == 1: + fold_params = dict(kernel_size=kernel_size, dilation=1, padding=0, stride=stride) + unfold = torch.nn.Unfold(**fold_params) + + fold_params2 = dict(kernel_size=(kernel_size[0] // df, kernel_size[0] // df), + dilation=1, padding=0, + stride=(stride[0] // df, stride[1] // df)) + fold = torch.nn.Fold(output_size=(x.shape[2] // df, x.shape[3] // df), **fold_params2) + + weighting = self.get_weighting(kernel_size[0] // df, kernel_size[1] // df, Ly, Lx, x.device).to(x.dtype) + normalization = fold(weighting).view(1, 1, h // df, w // df) # normalizes the overlap + weighting = weighting.view((1, 1, kernel_size[0] // df, kernel_size[1] // df, Ly * Lx)) + + else: + raise NotImplementedError + + return fold, unfold, normalization, weighting + + @torch.no_grad() + def get_input(self, batch, k, return_first_stage_outputs=False, force_c_encode=False, + cond_key=None, return_original_cond=False, bs=None): + x = super().get_input(batch, k) + if bs is not None: + x = x[:bs] + x = x.to(self.device) + encoder_posterior = self.encode_first_stage(x) + z = self.get_first_stage_encoding(encoder_posterior).detach() + + if self.model.conditioning_key is not None: + if cond_key is None: + cond_key = self.cond_stage_key + if cond_key != self.first_stage_key: + if cond_key in ['caption', 'coordinates_bbox']: + xc = batch[cond_key] + elif cond_key == 'class_label': + xc = batch + else: + xc = super().get_input(batch, cond_key).to(self.device) + else: + xc = x + if not self.cond_stage_trainable or force_c_encode: + if isinstance(xc, dict) or isinstance(xc, list): + # import pudb; pudb.set_trace() + c = self.get_learned_conditioning(xc) + else: + c = self.get_learned_conditioning(xc.to(self.device)) + else: + c = xc + if bs is not None: + c = c[:bs] + + if self.use_positional_encodings: + pos_x, pos_y = self.compute_latent_shifts(batch) + ckey = __conditioning_keys__[self.model.conditioning_key] + c = {ckey: c, 'pos_x': pos_x, 'pos_y': pos_y} + + else: + c = None + xc = None + if self.use_positional_encodings: + pos_x, pos_y = self.compute_latent_shifts(batch) + c = {'pos_x': pos_x, 'pos_y': pos_y} + out = [z, c] + if return_first_stage_outputs: + xrec = self.decode_first_stage(z) + out.extend([x, xrec]) + if return_original_cond: + out.append(xc) + return out + + @torch.no_grad() + def decode_first_stage(self, z, predict_cids=False, force_not_quantize=False): + if predict_cids: + if z.dim() == 4: + z = torch.argmax(z.exp(), dim=1).long() + z = self.first_stage_model.quantize.get_codebook_entry(z, shape=None) + z = rearrange(z, 'b h w c -> b c h w').contiguous() + + z = 1. / self.scale_factor * z + + if hasattr(self, "split_input_params"): + if self.split_input_params["patch_distributed_vq"]: + ks = self.split_input_params["ks"] # eg. (128, 128) + stride = self.split_input_params["stride"] # eg. (64, 64) + uf = self.split_input_params["vqf"] + bs, nc, h, w = z.shape + if ks[0] > h or ks[1] > w: + ks = (min(ks[0], h), min(ks[1], w)) + print("reducing Kernel") + + if stride[0] > h or stride[1] > w: + stride = (min(stride[0], h), min(stride[1], w)) + print("reducing stride") + + fold, unfold, normalization, weighting = self.get_fold_unfold(z, ks, stride, uf=uf) + + z = unfold(z) # (bn, nc * prod(**ks), L) + # 1. Reshape to img shape + z = z.view((z.shape[0], -1, ks[0], ks[1], z.shape[-1])) # (bn, nc, ks[0], ks[1], L ) + + # 2. apply model loop over last dim + if isinstance(self.first_stage_model, VQModelInterface): + output_list = [self.first_stage_model.decode(z[:, :, :, :, i], + force_not_quantize=predict_cids or force_not_quantize) + for i in range(z.shape[-1])] + else: + + output_list = [self.first_stage_model.decode(z[:, :, :, :, i]) + for i in range(z.shape[-1])] + + o = torch.stack(output_list, axis=-1) # # (bn, nc, ks[0], ks[1], L) + o = o * weighting + # Reverse 1. reshape to img shape + o = o.view((o.shape[0], -1, o.shape[-1])) # (bn, nc * ks[0] * ks[1], L) + # stitch crops together + decoded = fold(o) + decoded = decoded / normalization # norm is shape (1, 1, h, w) + return decoded + else: + if isinstance(self.first_stage_model, VQModelInterface): + return self.first_stage_model.decode(z, force_not_quantize=predict_cids or force_not_quantize) + else: + return self.first_stage_model.decode(z) + + else: + if isinstance(self.first_stage_model, VQModelInterface): + return self.first_stage_model.decode(z, force_not_quantize=predict_cids or force_not_quantize) + else: + return self.first_stage_model.decode(z) + + # same as above but without decorator + def differentiable_decode_first_stage(self, z, predict_cids=False, force_not_quantize=False): + if predict_cids: + if z.dim() == 4: + z = torch.argmax(z.exp(), dim=1).long() + z = self.first_stage_model.quantize.get_codebook_entry(z, shape=None) + z = rearrange(z, 'b h w c -> b c h w').contiguous() + + z = 1. / self.scale_factor * z + + if hasattr(self, "split_input_params"): + if self.split_input_params["patch_distributed_vq"]: + ks = self.split_input_params["ks"] # eg. (128, 128) + stride = self.split_input_params["stride"] # eg. (64, 64) + uf = self.split_input_params["vqf"] + bs, nc, h, w = z.shape + if ks[0] > h or ks[1] > w: + ks = (min(ks[0], h), min(ks[1], w)) + print("reducing Kernel") + + if stride[0] > h or stride[1] > w: + stride = (min(stride[0], h), min(stride[1], w)) + print("reducing stride") + + fold, unfold, normalization, weighting = self.get_fold_unfold(z, ks, stride, uf=uf) + + z = unfold(z) # (bn, nc * prod(**ks), L) + # 1. Reshape to img shape + z = z.view((z.shape[0], -1, ks[0], ks[1], z.shape[-1])) # (bn, nc, ks[0], ks[1], L ) + + # 2. apply model loop over last dim + if isinstance(self.first_stage_model, VQModelInterface): + output_list = [self.first_stage_model.decode(z[:, :, :, :, i], + force_not_quantize=predict_cids or force_not_quantize) + for i in range(z.shape[-1])] + else: + + output_list = [self.first_stage_model.decode(z[:, :, :, :, i]) + for i in range(z.shape[-1])] + + o = torch.stack(output_list, axis=-1) # # (bn, nc, ks[0], ks[1], L) + o = o * weighting + # Reverse 1. reshape to img shape + o = o.view((o.shape[0], -1, o.shape[-1])) # (bn, nc * ks[0] * ks[1], L) + # stitch crops together + decoded = fold(o) + decoded = decoded / normalization # norm is shape (1, 1, h, w) + return decoded + else: + if isinstance(self.first_stage_model, VQModelInterface): + return self.first_stage_model.decode(z, force_not_quantize=predict_cids or force_not_quantize) + else: + return self.first_stage_model.decode(z) + + else: + if isinstance(self.first_stage_model, VQModelInterface): + return self.first_stage_model.decode(z, force_not_quantize=predict_cids or force_not_quantize) + else: + return self.first_stage_model.decode(z) + + @torch.no_grad() + def encode_first_stage(self, x): + if hasattr(self, "split_input_params"): + if self.split_input_params["patch_distributed_vq"]: + ks = self.split_input_params["ks"] # eg. (128, 128) + stride = self.split_input_params["stride"] # eg. (64, 64) + df = self.split_input_params["vqf"] + self.split_input_params['original_image_size'] = x.shape[-2:] + bs, nc, h, w = x.shape + if ks[0] > h or ks[1] > w: + ks = (min(ks[0], h), min(ks[1], w)) + print("reducing Kernel") + + if stride[0] > h or stride[1] > w: + stride = (min(stride[0], h), min(stride[1], w)) + print("reducing stride") + + fold, unfold, normalization, weighting = self.get_fold_unfold(x, ks, stride, df=df) + z = unfold(x) # (bn, nc * prod(**ks), L) + # Reshape to img shape + z = z.view((z.shape[0], -1, ks[0], ks[1], z.shape[-1])) # (bn, nc, ks[0], ks[1], L ) + + output_list = [self.first_stage_model.encode(z[:, :, :, :, i]) + for i in range(z.shape[-1])] + + o = torch.stack(output_list, axis=-1) + o = o * weighting + + # Reverse reshape to img shape + o = o.view((o.shape[0], -1, o.shape[-1])) # (bn, nc * ks[0] * ks[1], L) + # stitch crops together + decoded = fold(o) + decoded = decoded / normalization + return decoded + + else: + return self.first_stage_model.encode(x) + else: + return self.first_stage_model.encode(x) + + def shared_step(self, batch, **kwargs): + x, c = self.get_input(batch, self.first_stage_key) + loss = self(x, c) + return loss + + def forward(self, x, c, *args, **kwargs): + t = torch.randint(0, self.num_timesteps, (x.shape[0],), device=self.device).long() + if self.model.conditioning_key is not None: + assert c is not None + if self.cond_stage_trainable: + c = self.get_learned_conditioning(c) + if self.shorten_cond_schedule: # TODO: drop this option + tc = self.cond_ids[t].to(self.device) + c = self.q_sample(x_start=c, t=tc, noise=torch.randn_like(c.float())) + return self.p_losses(x, c, t, *args, **kwargs) + + def apply_model(self, x_noisy, t, cond, return_ids=False): + + if isinstance(cond, dict): + # hybrid case, cond is exptected to be a dict + pass + else: + if not isinstance(cond, list): + cond = [cond] + key = 'c_concat' if self.model.conditioning_key == 'concat' else 'c_crossattn' + cond = {key: cond} + + if hasattr(self, "split_input_params"): + assert len(cond) == 1 # todo can only deal with one conditioning atm + assert not return_ids + ks = self.split_input_params["ks"] # eg. (128, 128) + stride = self.split_input_params["stride"] # eg. (64, 64) + + h, w = x_noisy.shape[-2:] + + fold, unfold, normalization, weighting = self.get_fold_unfold(x_noisy, ks, stride) + + z = unfold(x_noisy) # (bn, nc * prod(**ks), L) + # Reshape to img shape + z = z.view((z.shape[0], -1, ks[0], ks[1], z.shape[-1])) # (bn, nc, ks[0], ks[1], L ) + z_list = [z[:, :, :, :, i] for i in range(z.shape[-1])] + + if self.cond_stage_key in ["image", "LR_image", "segmentation", + 'bbox_img'] and self.model.conditioning_key: # todo check for completeness + c_key = next(iter(cond.keys())) # get key + c = next(iter(cond.values())) # get value + assert (len(c) == 1) # todo extend to list with more than one elem + c = c[0] # get element + + c = unfold(c) + c = c.view((c.shape[0], -1, ks[0], ks[1], c.shape[-1])) # (bn, nc, ks[0], ks[1], L ) + + cond_list = [{c_key: [c[:, :, :, :, i]]} for i in range(c.shape[-1])] + + elif self.cond_stage_key == 'coordinates_bbox': + assert 'original_image_size' in self.split_input_params, 'BoudingBoxRescaling is missing original_image_size' + + # assuming padding of unfold is always 0 and its dilation is always 1 + n_patches_per_row = int((w - ks[0]) / stride[0] + 1) + full_img_h, full_img_w = self.split_input_params['original_image_size'] + # as we are operating on latents, we need the factor from the original image size to the + # spatial latent size to properly rescale the crops for regenerating the bbox annotations + num_downs = self.first_stage_model.encoder.num_resolutions - 1 + rescale_latent = 2 ** (num_downs) + + # get top left postions of patches as conforming for the bbbox tokenizer, therefore we + # need to rescale the tl patch coordinates to be in between (0,1) + tl_patch_coordinates = [(rescale_latent * stride[0] * (patch_nr % n_patches_per_row) / full_img_w, + rescale_latent * stride[1] * (patch_nr // n_patches_per_row) / full_img_h) + for patch_nr in range(z.shape[-1])] + + # patch_limits are tl_coord, width and height coordinates as (x_tl, y_tl, h, w) + patch_limits = [(x_tl, y_tl, + rescale_latent * ks[0] / full_img_w, + rescale_latent * ks[1] / full_img_h) for x_tl, y_tl in tl_patch_coordinates] + # patch_values = [(np.arange(x_tl,min(x_tl+ks, 1.)),np.arange(y_tl,min(y_tl+ks, 1.))) for x_tl, y_tl in tl_patch_coordinates] + + # tokenize crop coordinates for the bounding boxes of the respective patches + patch_limits_tknzd = [torch.LongTensor(self.bbox_tokenizer._crop_encoder(bbox))[None].to(self.device) + for bbox in patch_limits] # list of length l with tensors of shape (1, 2) + print(patch_limits_tknzd[0].shape) + # cut tknzd crop position from conditioning + assert isinstance(cond, dict), 'cond must be dict to be fed into model' + cut_cond = cond['c_crossattn'][0][..., :-2].to(self.device) + print(cut_cond.shape) + + adapted_cond = torch.stack([torch.cat([cut_cond, p], dim=1) for p in patch_limits_tknzd]) + adapted_cond = rearrange(adapted_cond, 'l b n -> (l b) n') + print(adapted_cond.shape) + adapted_cond = self.get_learned_conditioning(adapted_cond) + print(adapted_cond.shape) + adapted_cond = rearrange(adapted_cond, '(l b) n d -> l b n d', l=z.shape[-1]) + print(adapted_cond.shape) + + cond_list = [{'c_crossattn': [e]} for e in adapted_cond] + + else: + cond_list = [cond for i in range(z.shape[-1])] # Todo make this more efficient + + # apply model by loop over crops + output_list = [self.model(z_list[i], t, **cond_list[i]) for i in range(z.shape[-1])] + assert not isinstance(output_list[0], + tuple) # todo cant deal with multiple model outputs check this never happens + + o = torch.stack(output_list, axis=-1) + o = o * weighting + # Reverse reshape to img shape + o = o.view((o.shape[0], -1, o.shape[-1])) # (bn, nc * ks[0] * ks[1], L) + # stitch crops together + x_recon = fold(o) / normalization + + else: + x_recon = self.model(x_noisy, t, **cond) + + if isinstance(x_recon, tuple) and not return_ids: + return x_recon[0] + else: + return x_recon + + def _predict_eps_from_xstart(self, x_t, t, pred_xstart): + return (extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t - pred_xstart) / \ + extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) + + def _prior_bpd(self, x_start): + """ + Get the prior KL term for the variational lower-bound, measured in + bits-per-dim. + This term can't be optimized, as it only depends on the encoder. + :param x_start: the [N x C x ...] tensor of inputs. + :return: a batch of [N] KL values (in bits), one per batch element. + """ + batch_size = x_start.shape[0] + t = torch.tensor([self.num_timesteps - 1] * batch_size, device=x_start.device) + qt_mean, _, qt_log_variance = self.q_mean_variance(x_start, t) + kl_prior = normal_kl(mean1=qt_mean, logvar1=qt_log_variance, mean2=0.0, logvar2=0.0) + return mean_flat(kl_prior) / np.log(2.0) + + def p_losses(self, x_start, cond, t, noise=None): + noise = default(noise, lambda: torch.randn_like(x_start)) + x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise) + model_output = self.apply_model(x_noisy, t, cond) + + loss_dict = {} + prefix = 'train' if self.training else 'val' + + if self.parameterization == "x0": + target = x_start + elif self.parameterization == "eps": + target = noise + else: + raise NotImplementedError() + + loss_simple = self.get_loss(model_output, target, mean=False).mean([1, 2, 3]) + loss_dict.update({f'{prefix}/loss_simple': loss_simple.mean()}) + + logvar_t = self.logvar[t].to(self.device) + loss = loss_simple / torch.exp(logvar_t) + logvar_t + # loss = loss_simple / torch.exp(self.logvar) + self.logvar + if self.learn_logvar: + loss_dict.update({f'{prefix}/loss_gamma': loss.mean()}) + loss_dict.update({'logvar': self.logvar.data.mean()}) + + loss = self.l_simple_weight * loss.mean() + + loss_vlb = self.get_loss(model_output, target, mean=False).mean(dim=(1, 2, 3)) + loss_vlb = (self.lvlb_weights[t] * loss_vlb).mean() + loss_dict.update({f'{prefix}/loss_vlb': loss_vlb}) + loss += (self.original_elbo_weight * loss_vlb) + loss_dict.update({f'{prefix}/loss': loss}) + + return loss, loss_dict + + def p_mean_variance(self, x, c, t, clip_denoised: bool, return_codebook_ids=False, quantize_denoised=False, + return_x0=False, score_corrector=None, corrector_kwargs=None): + t_in = t + model_out = self.apply_model(x, t_in, c, return_ids=return_codebook_ids) + + if score_corrector is not None: + assert self.parameterization == "eps" + model_out = score_corrector.modify_score(self, model_out, x, t, c, **corrector_kwargs) + + if return_codebook_ids: + model_out, logits = model_out + + if self.parameterization == "eps": + x_recon = self.predict_start_from_noise(x, t=t, noise=model_out) + elif self.parameterization == "x0": + x_recon = model_out + else: + raise NotImplementedError() + + if clip_denoised: + x_recon.clamp_(-1., 1.) + if quantize_denoised: + x_recon, _, [_, _, indices] = self.first_stage_model.quantize(x_recon) + model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t) + if return_codebook_ids: + return model_mean, posterior_variance, posterior_log_variance, logits + elif return_x0: + return model_mean, posterior_variance, posterior_log_variance, x_recon + else: + return model_mean, posterior_variance, posterior_log_variance + + @torch.no_grad() + def p_sample(self, x, c, t, clip_denoised=False, repeat_noise=False, + return_codebook_ids=False, quantize_denoised=False, return_x0=False, + temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None): + b, *_, device = *x.shape, x.device + outputs = self.p_mean_variance(x=x, c=c, t=t, clip_denoised=clip_denoised, + return_codebook_ids=return_codebook_ids, + quantize_denoised=quantize_denoised, + return_x0=return_x0, + score_corrector=score_corrector, corrector_kwargs=corrector_kwargs) + if return_codebook_ids: + raise DeprecationWarning("Support dropped.") + model_mean, _, model_log_variance, logits = outputs + elif return_x0: + model_mean, _, model_log_variance, x0 = outputs + else: + model_mean, _, model_log_variance = outputs + + noise = noise_like(x.shape, device, repeat_noise) * temperature + if noise_dropout > 0.: + noise = torch.nn.functional.dropout(noise, p=noise_dropout) + # no noise when t == 0 + nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1))) + + if return_codebook_ids: + return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise, logits.argmax(dim=1) + if return_x0: + return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise, x0 + else: + return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise + + @torch.no_grad() + def progressive_denoising(self, cond, shape, verbose=True, callback=None, quantize_denoised=False, + img_callback=None, mask=None, x0=None, temperature=1., noise_dropout=0., + score_corrector=None, corrector_kwargs=None, batch_size=None, x_T=None, start_T=None, + log_every_t=None): + if not log_every_t: + log_every_t = self.log_every_t + timesteps = self.num_timesteps + if batch_size is not None: + b = batch_size if batch_size is not None else shape[0] + shape = [batch_size] + list(shape) + else: + b = batch_size = shape[0] + if x_T is None: + img = torch.randn(shape, device=self.device) + else: + img = x_T + intermediates = [] + if cond is not None: + if isinstance(cond, dict): + cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else + [x[:batch_size] for x in cond[key]] for key in cond} + else: + cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size] + + if start_T is not None: + timesteps = min(timesteps, start_T) + iterator = tqdm(reversed(range(0, timesteps)), desc='Progressive Generation', + total=timesteps) if verbose else reversed( + range(0, timesteps)) + if type(temperature) == float: + temperature = [temperature] * timesteps + + for i in iterator: + ts = torch.full((b,), i, device=self.device, dtype=torch.long) + if self.shorten_cond_schedule: + assert self.model.conditioning_key != 'hybrid' + tc = self.cond_ids[ts].to(cond.device) + cond = self.q_sample(x_start=cond, t=tc, noise=torch.randn_like(cond)) + + img, x0_partial = self.p_sample(img, cond, ts, + clip_denoised=self.clip_denoised, + quantize_denoised=quantize_denoised, return_x0=True, + temperature=temperature[i], noise_dropout=noise_dropout, + score_corrector=score_corrector, corrector_kwargs=corrector_kwargs) + if mask is not None: + assert x0 is not None + img_orig = self.q_sample(x0, ts) + img = img_orig * mask + (1. - mask) * img + + if i % log_every_t == 0 or i == timesteps - 1: + intermediates.append(x0_partial) + if callback: + callback(i) + if img_callback: + img_callback(img, i) + return img, intermediates + + @torch.no_grad() + def p_sample_loop(self, cond, shape, return_intermediates=False, + x_T=None, verbose=True, callback=None, timesteps=None, quantize_denoised=False, + mask=None, x0=None, img_callback=None, start_T=None, + log_every_t=None): + + if not log_every_t: + log_every_t = self.log_every_t + device = self.betas.device + b = shape[0] + if x_T is None: + img = torch.randn(shape, device=device) + else: + img = x_T + + intermediates = [img] + if timesteps is None: + timesteps = self.num_timesteps + + if start_T is not None: + timesteps = min(timesteps, start_T) + iterator = tqdm(reversed(range(0, timesteps)), desc='Sampling t', total=timesteps) if verbose else reversed( + range(0, timesteps)) + + if mask is not None: + assert x0 is not None + assert x0.shape[2:3] == mask.shape[2:3] # spatial size has to match + + for i in iterator: + ts = torch.full((b,), i, device=device, dtype=torch.long) + if self.shorten_cond_schedule: + assert self.model.conditioning_key != 'hybrid' + tc = self.cond_ids[ts].to(cond.device) + cond = self.q_sample(x_start=cond, t=tc, noise=torch.randn_like(cond)) + + img = self.p_sample(img, cond, ts, + clip_denoised=self.clip_denoised, + quantize_denoised=quantize_denoised) + if mask is not None: + img_orig = self.q_sample(x0, ts) + img = img_orig * mask + (1. - mask) * img + + if i % log_every_t == 0 or i == timesteps - 1: + intermediates.append(img) + if callback: + callback(i) + if img_callback: + img_callback(img, i) + + if return_intermediates: + return img, intermediates + return img + + @torch.no_grad() + def sample(self, cond, batch_size=16, return_intermediates=False, x_T=None, + verbose=True, timesteps=None, quantize_denoised=False, + mask=None, x0=None, shape=None,**kwargs): + if shape is None: + shape = (batch_size, self.channels, self.image_size, self.image_size) + if cond is not None: + if isinstance(cond, dict): + cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else + [x[:batch_size] for x in cond[key]] for key in cond} + else: + cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size] + return self.p_sample_loop(cond, + shape, + return_intermediates=return_intermediates, x_T=x_T, + verbose=verbose, timesteps=timesteps, quantize_denoised=quantize_denoised, + mask=mask, x0=x0) + + @torch.no_grad() + def sample_log(self,cond,batch_size,ddim, ddim_steps,**kwargs): + + if ddim: + ddim_sampler = DDIMSampler(self) + shape = (self.channels, self.image_size, self.image_size) + samples, intermediates =ddim_sampler.sample(ddim_steps,batch_size, + shape,cond,verbose=False,**kwargs) + + else: + samples, intermediates = self.sample(cond=cond, batch_size=batch_size, + return_intermediates=True,**kwargs) + + return samples, intermediates + + + @torch.no_grad() + def log_images(self, batch, N=8, n_row=4, sample=True, ddim_steps=200, ddim_eta=1., return_keys=None, + quantize_denoised=True, inpaint=True, plot_denoise_rows=False, plot_progressive_rows=True, + plot_diffusion_rows=True, **kwargs): + + use_ddim = ddim_steps is not None + + log = {} + z, c, x, xrec, xc = self.get_input(batch, self.first_stage_key, + return_first_stage_outputs=True, + force_c_encode=True, + return_original_cond=True, + bs=N) + N = min(x.shape[0], N) + n_row = min(x.shape[0], n_row) + log["inputs"] = x + log["reconstruction"] = xrec + if self.model.conditioning_key is not None: + if hasattr(self.cond_stage_model, "decode"): + xc = self.cond_stage_model.decode(c) + log["conditioning"] = xc + elif self.cond_stage_key in ["caption"]: + xc = log_txt_as_img((x.shape[2], x.shape[3]), batch["caption"]) + log["conditioning"] = xc + elif self.cond_stage_key == 'class_label': + xc = log_txt_as_img((x.shape[2], x.shape[3]), batch["human_label"]) + log['conditioning'] = xc + elif isimage(xc): + log["conditioning"] = xc + if ismap(xc): + log["original_conditioning"] = self.to_rgb(xc) + + if plot_diffusion_rows: + # get diffusion row + diffusion_row = [] + z_start = z[:n_row] + for t in range(self.num_timesteps): + if t % self.log_every_t == 0 or t == self.num_timesteps - 1: + t = repeat(torch.tensor([t]), '1 -> b', b=n_row) + t = t.to(self.device).long() + noise = torch.randn_like(z_start) + z_noisy = self.q_sample(x_start=z_start, t=t, noise=noise) + diffusion_row.append(self.decode_first_stage(z_noisy)) + + diffusion_row = torch.stack(diffusion_row) # n_log_step, n_row, C, H, W + diffusion_grid = rearrange(diffusion_row, 'n b c h w -> b n c h w') + diffusion_grid = rearrange(diffusion_grid, 'b n c h w -> (b n) c h w') + diffusion_grid = make_grid(diffusion_grid, nrow=diffusion_row.shape[0]) + log["diffusion_row"] = diffusion_grid + + if sample: + # get denoise row + with self.ema_scope("Plotting"): + samples, z_denoise_row = self.sample_log(cond=c,batch_size=N,ddim=use_ddim, + ddim_steps=ddim_steps,eta=ddim_eta) + # samples, z_denoise_row = self.sample(cond=c, batch_size=N, return_intermediates=True) + x_samples = self.decode_first_stage(samples) + log["samples"] = x_samples + if plot_denoise_rows: + denoise_grid = self._get_denoise_row_from_list(z_denoise_row) + log["denoise_row"] = denoise_grid + + if quantize_denoised and not isinstance(self.first_stage_model, AutoencoderKL) and not isinstance( + self.first_stage_model, IdentityFirstStage): + # also display when quantizing x0 while sampling + with self.ema_scope("Plotting Quantized Denoised"): + samples, z_denoise_row = self.sample_log(cond=c,batch_size=N,ddim=use_ddim, + ddim_steps=ddim_steps,eta=ddim_eta, + quantize_denoised=True) + # samples, z_denoise_row = self.sample(cond=c, batch_size=N, return_intermediates=True, + # quantize_denoised=True) + x_samples = self.decode_first_stage(samples.to(self.device)) + log["samples_x0_quantized"] = x_samples + + if inpaint: + # make a simple center square + h, w = z.shape[2], z.shape[3] + mask = torch.ones(N, h, w).to(self.device) + # zeros will be filled in + mask[:, h // 4:3 * h // 4, w // 4:3 * w // 4] = 0. + mask = mask[:, None, ...] + with self.ema_scope("Plotting Inpaint"): + + samples, _ = self.sample_log(cond=c,batch_size=N,ddim=use_ddim, eta=ddim_eta, + ddim_steps=ddim_steps, x0=z[:N], mask=mask) + x_samples = self.decode_first_stage(samples.to(self.device)) + log["samples_inpainting"] = x_samples + log["mask"] = mask + + # outpaint + with self.ema_scope("Plotting Outpaint"): + samples, _ = self.sample_log(cond=c, batch_size=N, ddim=use_ddim,eta=ddim_eta, + ddim_steps=ddim_steps, x0=z[:N], mask=mask) + x_samples = self.decode_first_stage(samples.to(self.device)) + log["samples_outpainting"] = x_samples + + if plot_progressive_rows: + with self.ema_scope("Plotting Progressives"): + img, progressives = self.progressive_denoising(c, + shape=(self.channels, self.image_size, self.image_size), + batch_size=N) + prog_row = self._get_denoise_row_from_list(progressives, desc="Progressive Generation") + log["progressive_row"] = prog_row + + if return_keys: + if np.intersect1d(list(log.keys()), return_keys).shape[0] == 0: + return log + else: + return {key: log[key] for key in return_keys} + return log + + def configure_optimizers(self): + lr = self.learning_rate + params = list(self.model.parameters()) + if self.cond_stage_trainable: + print(f"{self.__class__.__name__}: Also optimizing conditioner params!") + params = params + list(self.cond_stage_model.parameters()) + if self.learn_logvar: + print('Diffusion model optimizing logvar') + params.append(self.logvar) + opt = torch.optim.AdamW(params, lr=lr) + if self.use_scheduler: + assert 'target' in self.scheduler_config + scheduler = instantiate_from_config(self.scheduler_config) + + print("Setting up LambdaLR scheduler...") + scheduler = [ + { + 'scheduler': LambdaLR(opt, lr_lambda=scheduler.schedule), + 'interval': 'step', + 'frequency': 1 + }] + return [opt], scheduler + return opt + + @torch.no_grad() + def to_rgb(self, x): + x = x.float() + if not hasattr(self, "colorize"): + self.colorize = torch.randn(3, x.shape[1], 1, 1).to(x) + x = nn.functional.conv2d(x, weight=self.colorize) + x = 2. * (x - x.min()) / (x.max() - x.min()) - 1. + return x + + +class DiffusionWrapperV1(pl.LightningModule): + def __init__(self, diff_model_config, conditioning_key): + super().__init__() + self.diffusion_model = instantiate_from_config(diff_model_config) + self.conditioning_key = conditioning_key + assert self.conditioning_key in [None, 'concat', 'crossattn', 'hybrid', 'adm'] + + def forward(self, x, t, c_concat: list = None, c_crossattn: list = None): + if self.conditioning_key is None: + out = self.diffusion_model(x, t) + elif self.conditioning_key == 'concat': + xc = torch.cat([x] + c_concat, dim=1) + out = self.diffusion_model(xc, t) + elif self.conditioning_key == 'crossattn': + cc = torch.cat(c_crossattn, 1) + out = self.diffusion_model(x, t, context=cc) + elif self.conditioning_key == 'hybrid': + xc = torch.cat([x] + c_concat, dim=1) + cc = torch.cat(c_crossattn, 1) + out = self.diffusion_model(xc, t, context=cc) + elif self.conditioning_key == 'adm': + cc = c_crossattn[0] + out = self.diffusion_model(x, t, y=cc) + else: + raise NotImplementedError() + + return out + + +class Layout2ImgDiffusionV1(LatentDiffusionV1): + # TODO: move all layout-specific hacks to this class + def __init__(self, cond_stage_key, *args, **kwargs): + assert cond_stage_key == 'coordinates_bbox', 'Layout2ImgDiffusion only for cond_stage_key="coordinates_bbox"' + super().__init__(*args, cond_stage_key=cond_stage_key, **kwargs) + + def log_images(self, batch, N=8, *args, **kwargs): + logs = super().log_images(*args, batch=batch, N=N, **kwargs) + + key = 'train' if self.training else 'validation' + dset = self.trainer.datamodule.datasets[key] + mapper = dset.conditional_builders[self.cond_stage_key] + + bbox_imgs = [] + map_fn = lambda catno: dset.get_textual_label(dset.get_category_id(catno)) + for tknzd_bbox in batch[self.cond_stage_key][:N]: + bboximg = mapper.plot(tknzd_bbox.detach().cpu(), map_fn, (256, 256)) + bbox_imgs.append(bboximg) + + cond_img = torch.stack(bbox_imgs, dim=0) + logs['bbox_image'] = cond_img + return logs + +ldm.models.diffusion.ddpm.DDPMV1 = DDPMV1 +ldm.models.diffusion.ddpm.LatentDiffusionV1 = LatentDiffusionV1 +ldm.models.diffusion.ddpm.DiffusionWrapperV1 = DiffusionWrapperV1 +ldm.models.diffusion.ddpm.Layout2ImgDiffusionV1 = Layout2ImgDiffusionV1 diff --git a/extensions-builtin/Lora/extra_networks_lora.py b/extensions-builtin/Lora/extra_networks_lora.py new file mode 100644 index 0000000000000000000000000000000000000000..3e4088e0a952647f74bfd93cf5489079fa8bde0c --- /dev/null +++ b/extensions-builtin/Lora/extra_networks_lora.py @@ -0,0 +1,45 @@ +from modules import extra_networks, shared +import lora + + +class ExtraNetworkLora(extra_networks.ExtraNetwork): + def __init__(self): + super().__init__('lora') + + def activate(self, p, params_list): + additional = shared.opts.sd_lora + + if additional != "None" and additional in lora.available_loras and len([x for x in params_list if x.items[0] == additional]) == 0: + p.all_prompts = [x + f"" for x in p.all_prompts] + params_list.append(extra_networks.ExtraNetworkParams(items=[additional, shared.opts.extra_networks_default_multiplier])) + + names = [] + multipliers = [] + for params in params_list: + assert len(params.items) > 0 + + names.append(params.items[0]) + multipliers.append(float(params.items[1]) if len(params.items) > 1 else 1.0) + + lora.load_loras(names, multipliers) + + if shared.opts.lora_add_hashes_to_infotext: + lora_hashes = [] + for item in lora.loaded_loras: + shorthash = item.lora_on_disk.shorthash + if not shorthash: + continue + + alias = item.mentioned_name + if not alias: + continue + + alias = alias.replace(":", "").replace(",", "") + + lora_hashes.append(f"{alias}: {shorthash}") + + if lora_hashes: + p.extra_generation_params["Lora hashes"] = ", ".join(lora_hashes) + + def deactivate(self, p): + pass diff --git a/extensions-builtin/Lora/lora.py b/extensions-builtin/Lora/lora.py new file mode 100644 index 0000000000000000000000000000000000000000..e31a17ca8e45a2f741dcbf9060fcb3f0c163ba09 --- /dev/null +++ b/extensions-builtin/Lora/lora.py @@ -0,0 +1,502 @@ +import os +import re +import torch +from typing import Union + +from modules import shared, devices, sd_models, errors, scripts, sd_hijack, hashes + +metadata_tags_order = {"ss_sd_model_name": 1, "ss_resolution": 2, "ss_clip_skip": 3, "ss_num_train_images": 10, "ss_tag_frequency": 20} + +re_digits = re.compile(r"\d+") +re_x_proj = re.compile(r"(.*)_([qkv]_proj)$") +re_compiled = {} + +suffix_conversion = { + "attentions": {}, + "resnets": { + "conv1": "in_layers_2", + "conv2": "out_layers_3", + "time_emb_proj": "emb_layers_1", + "conv_shortcut": "skip_connection", + } +} + + +def convert_diffusers_name_to_compvis(key, is_sd2): + def match(match_list, regex_text): + regex = re_compiled.get(regex_text) + if regex is None: + regex = re.compile(regex_text) + re_compiled[regex_text] = regex + + r = re.match(regex, key) + if not r: + return False + + match_list.clear() + match_list.extend([int(x) if re.match(re_digits, x) else x for x in r.groups()]) + return True + + m = [] + + if match(m, r"lora_unet_down_blocks_(\d+)_(attentions|resnets)_(\d+)_(.+)"): + suffix = suffix_conversion.get(m[1], {}).get(m[3], m[3]) + return f"diffusion_model_input_blocks_{1 + m[0] * 3 + m[2]}_{1 if m[1] == 'attentions' else 0}_{suffix}" + + if match(m, r"lora_unet_mid_block_(attentions|resnets)_(\d+)_(.+)"): + suffix = suffix_conversion.get(m[0], {}).get(m[2], m[2]) + return f"diffusion_model_middle_block_{1 if m[0] == 'attentions' else m[1] * 2}_{suffix}" + + if match(m, r"lora_unet_up_blocks_(\d+)_(attentions|resnets)_(\d+)_(.+)"): + suffix = suffix_conversion.get(m[1], {}).get(m[3], m[3]) + return f"diffusion_model_output_blocks_{m[0] * 3 + m[2]}_{1 if m[1] == 'attentions' else 0}_{suffix}" + + if match(m, r"lora_unet_down_blocks_(\d+)_downsamplers_0_conv"): + return f"diffusion_model_input_blocks_{3 + m[0] * 3}_0_op" + + if match(m, r"lora_unet_up_blocks_(\d+)_upsamplers_0_conv"): + return f"diffusion_model_output_blocks_{2 + m[0] * 3}_{2 if m[0]>0 else 1}_conv" + + if match(m, r"lora_te_text_model_encoder_layers_(\d+)_(.+)"): + if is_sd2: + if 'mlp_fc1' in m[1]: + return f"model_transformer_resblocks_{m[0]}_{m[1].replace('mlp_fc1', 'mlp_c_fc')}" + elif 'mlp_fc2' in m[1]: + return f"model_transformer_resblocks_{m[0]}_{m[1].replace('mlp_fc2', 'mlp_c_proj')}" + else: + return f"model_transformer_resblocks_{m[0]}_{m[1].replace('self_attn', 'attn')}" + + return f"transformer_text_model_encoder_layers_{m[0]}_{m[1]}" + + return key + + +class LoraOnDisk: + def __init__(self, name, filename): + self.name = name + self.filename = filename + self.metadata = {} + self.is_safetensors = os.path.splitext(filename)[1].lower() == ".safetensors" + + if self.is_safetensors: + try: + self.metadata = sd_models.read_metadata_from_safetensors(filename) + except Exception as e: + errors.display(e, f"reading lora {filename}") + + if self.metadata: + m = {} + for k, v in sorted(self.metadata.items(), key=lambda x: metadata_tags_order.get(x[0], 999)): + m[k] = v + + self.metadata = m + + self.ssmd_cover_images = self.metadata.pop('ssmd_cover_images', None) # those are cover images and they are too big to display in UI as text + self.alias = self.metadata.get('ss_output_name', self.name) + + self.hash = None + self.shorthash = None + self.set_hash( + self.metadata.get('sshs_model_hash') or + hashes.sha256_from_cache(self.filename, "lora/" + self.name, use_addnet_hash=self.is_safetensors) or + '' + ) + + def set_hash(self, v): + self.hash = v + self.shorthash = self.hash[0:12] + + if self.shorthash: + available_lora_hash_lookup[self.shorthash] = self + + def read_hash(self): + if not self.hash: + self.set_hash(hashes.sha256(self.filename, "lora/" + self.name, use_addnet_hash=self.is_safetensors) or '') + + def get_alias(self): + if shared.opts.lora_preferred_name == "Filename" or self.alias.lower() in forbidden_lora_aliases: + return self.name + else: + return self.alias + + +class LoraModule: + def __init__(self, name, lora_on_disk: LoraOnDisk): + self.name = name + self.lora_on_disk = lora_on_disk + self.multiplier = 1.0 + self.modules = {} + self.mtime = None + + self.mentioned_name = None + """the text that was used to add lora to prompt - can be either name or an alias""" + + +class LoraUpDownModule: + def __init__(self): + self.up = None + self.down = None + self.alpha = None + + +def assign_lora_names_to_compvis_modules(sd_model): + lora_layer_mapping = {} + + for name, module in shared.sd_model.cond_stage_model.wrapped.named_modules(): + lora_name = name.replace(".", "_") + lora_layer_mapping[lora_name] = module + module.lora_layer_name = lora_name + + for name, module in shared.sd_model.model.named_modules(): + lora_name = name.replace(".", "_") + lora_layer_mapping[lora_name] = module + module.lora_layer_name = lora_name + + sd_model.lora_layer_mapping = lora_layer_mapping + + +def load_lora(name, lora_on_disk): + lora = LoraModule(name, lora_on_disk) + lora.mtime = os.path.getmtime(lora_on_disk.filename) + + sd = sd_models.read_state_dict(lora_on_disk.filename) + + # this should not be needed but is here as an emergency fix for an unknown error people are experiencing in 1.2.0 + if not hasattr(shared.sd_model, 'lora_layer_mapping'): + assign_lora_names_to_compvis_modules(shared.sd_model) + + keys_failed_to_match = {} + is_sd2 = 'model_transformer_resblocks' in shared.sd_model.lora_layer_mapping + + for key_diffusers, weight in sd.items(): + key_diffusers_without_lora_parts, lora_key = key_diffusers.split(".", 1) + key = convert_diffusers_name_to_compvis(key_diffusers_without_lora_parts, is_sd2) + + sd_module = shared.sd_model.lora_layer_mapping.get(key, None) + + if sd_module is None: + m = re_x_proj.match(key) + if m: + sd_module = shared.sd_model.lora_layer_mapping.get(m.group(1), None) + + if sd_module is None: + keys_failed_to_match[key_diffusers] = key + continue + + lora_module = lora.modules.get(key, None) + if lora_module is None: + lora_module = LoraUpDownModule() + lora.modules[key] = lora_module + + if lora_key == "alpha": + lora_module.alpha = weight.item() + continue + + if type(sd_module) == torch.nn.Linear: + module = torch.nn.Linear(weight.shape[1], weight.shape[0], bias=False) + elif type(sd_module) == torch.nn.modules.linear.NonDynamicallyQuantizableLinear: + module = torch.nn.Linear(weight.shape[1], weight.shape[0], bias=False) + elif type(sd_module) == torch.nn.MultiheadAttention: + module = torch.nn.Linear(weight.shape[1], weight.shape[0], bias=False) + elif type(sd_module) == torch.nn.Conv2d and weight.shape[2:] == (1, 1): + module = torch.nn.Conv2d(weight.shape[1], weight.shape[0], (1, 1), bias=False) + elif type(sd_module) == torch.nn.Conv2d and weight.shape[2:] == (3, 3): + module = torch.nn.Conv2d(weight.shape[1], weight.shape[0], (3, 3), bias=False) + else: + print(f'Lora layer {key_diffusers} matched a layer with unsupported type: {type(sd_module).__name__}') + continue + raise AssertionError(f"Lora layer {key_diffusers} matched a layer with unsupported type: {type(sd_module).__name__}") + + with torch.no_grad(): + module.weight.copy_(weight) + + module.to(device=devices.cpu, dtype=devices.dtype) + + if lora_key == "lora_up.weight": + lora_module.up = module + elif lora_key == "lora_down.weight": + lora_module.down = module + else: + raise AssertionError(f"Bad Lora layer name: {key_diffusers} - must end in lora_up.weight, lora_down.weight or alpha") + + if len(keys_failed_to_match) > 0: + print(f"Failed to match keys when loading Lora {lora_on_disk.filename}: {keys_failed_to_match}") + + return lora + + +def load_loras(names, multipliers=None): + already_loaded = {} + + for lora in loaded_loras: + if lora.name in names: + already_loaded[lora.name] = lora + + loaded_loras.clear() + + loras_on_disk = [available_lora_aliases.get(name, None) for name in names] + if any(x is None for x in loras_on_disk): + list_available_loras() + + loras_on_disk = [available_lora_aliases.get(name, None) for name in names] + + failed_to_load_loras = [] + + for i, name in enumerate(names): + lora = already_loaded.get(name, None) + + lora_on_disk = loras_on_disk[i] + + if lora_on_disk is not None: + if lora is None or os.path.getmtime(lora_on_disk.filename) > lora.mtime: + try: + lora = load_lora(name, lora_on_disk) + except Exception as e: + errors.display(e, f"loading Lora {lora_on_disk.filename}") + continue + + lora.mentioned_name = name + + lora_on_disk.read_hash() + + if lora is None: + failed_to_load_loras.append(name) + print(f"Couldn't find Lora with name {name}") + continue + + lora.multiplier = multipliers[i] if multipliers else 1.0 + loaded_loras.append(lora) + + if len(failed_to_load_loras) > 0: + sd_hijack.model_hijack.comments.append("Failed to find Loras: " + ", ".join(failed_to_load_loras)) + + +def lora_calc_updown(lora, module, target): + with torch.no_grad(): + up = module.up.weight.to(target.device, dtype=target.dtype) + down = module.down.weight.to(target.device, dtype=target.dtype) + + if up.shape[2:] == (1, 1) and down.shape[2:] == (1, 1): + updown = (up.squeeze(2).squeeze(2) @ down.squeeze(2).squeeze(2)).unsqueeze(2).unsqueeze(3) + elif up.shape[2:] == (3, 3) or down.shape[2:] == (3, 3): + updown = torch.nn.functional.conv2d(down.permute(1, 0, 2, 3), up).permute(1, 0, 2, 3) + else: + updown = up @ down + + updown = updown * lora.multiplier * (module.alpha / module.up.weight.shape[1] if module.alpha else 1.0) + + return updown + + +def lora_restore_weights_from_backup(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn.MultiheadAttention]): + weights_backup = getattr(self, "lora_weights_backup", None) + + if weights_backup is None: + return + + if isinstance(self, torch.nn.MultiheadAttention): + self.in_proj_weight.copy_(weights_backup[0]) + self.out_proj.weight.copy_(weights_backup[1]) + else: + self.weight.copy_(weights_backup) + + +def lora_apply_weights(self: Union[torch.nn.Conv2d, torch.nn.Linear, torch.nn.MultiheadAttention]): + """ + Applies the currently selected set of Loras to the weights of torch layer self. + If weights already have this particular set of loras applied, does nothing. + If not, restores orginal weights from backup and alters weights according to loras. + """ + + lora_layer_name = getattr(self, 'lora_layer_name', None) + if lora_layer_name is None: + return + + current_names = getattr(self, "lora_current_names", ()) + wanted_names = tuple((x.name, x.multiplier) for x in loaded_loras) + + weights_backup = getattr(self, "lora_weights_backup", None) + if weights_backup is None: + if isinstance(self, torch.nn.MultiheadAttention): + weights_backup = (self.in_proj_weight.to(devices.cpu, copy=True), self.out_proj.weight.to(devices.cpu, copy=True)) + else: + weights_backup = self.weight.to(devices.cpu, copy=True) + + self.lora_weights_backup = weights_backup + + if current_names != wanted_names: + lora_restore_weights_from_backup(self) + + for lora in loaded_loras: + module = lora.modules.get(lora_layer_name, None) + if module is not None and hasattr(self, 'weight'): + self.weight += lora_calc_updown(lora, module, self.weight) + continue + + module_q = lora.modules.get(lora_layer_name + "_q_proj", None) + module_k = lora.modules.get(lora_layer_name + "_k_proj", None) + module_v = lora.modules.get(lora_layer_name + "_v_proj", None) + module_out = lora.modules.get(lora_layer_name + "_out_proj", None) + + if isinstance(self, torch.nn.MultiheadAttention) and module_q and module_k and module_v and module_out: + updown_q = lora_calc_updown(lora, module_q, self.in_proj_weight) + updown_k = lora_calc_updown(lora, module_k, self.in_proj_weight) + updown_v = lora_calc_updown(lora, module_v, self.in_proj_weight) + updown_qkv = torch.vstack([updown_q, updown_k, updown_v]) + + self.in_proj_weight += updown_qkv + self.out_proj.weight += lora_calc_updown(lora, module_out, self.out_proj.weight) + continue + + if module is None: + continue + + print(f'failed to calculate lora weights for layer {lora_layer_name}') + + self.lora_current_names = wanted_names + + +def lora_forward(module, input, original_forward): + """ + Old way of applying Lora by executing operations during layer's forward. + Stacking many loras this way results in big performance degradation. + """ + + if len(loaded_loras) == 0: + return original_forward(module, input) + + input = devices.cond_cast_unet(input) + + lora_restore_weights_from_backup(module) + lora_reset_cached_weight(module) + + res = original_forward(module, input) + + lora_layer_name = getattr(module, 'lora_layer_name', None) + for lora in loaded_loras: + module = lora.modules.get(lora_layer_name, None) + if module is None: + continue + + module.up.to(device=devices.device) + module.down.to(device=devices.device) + + res = res + module.up(module.down(input)) * lora.multiplier * (module.alpha / module.up.weight.shape[1] if module.alpha else 1.0) + + return res + + +def lora_reset_cached_weight(self: Union[torch.nn.Conv2d, torch.nn.Linear]): + self.lora_current_names = () + self.lora_weights_backup = None + + +def lora_Linear_forward(self, input): + if shared.opts.lora_functional: + return lora_forward(self, input, torch.nn.Linear_forward_before_lora) + + lora_apply_weights(self) + + return torch.nn.Linear_forward_before_lora(self, input) + + +def lora_Linear_load_state_dict(self, *args, **kwargs): + lora_reset_cached_weight(self) + + return torch.nn.Linear_load_state_dict_before_lora(self, *args, **kwargs) + + +def lora_Conv2d_forward(self, input): + if shared.opts.lora_functional: + return lora_forward(self, input, torch.nn.Conv2d_forward_before_lora) + + lora_apply_weights(self) + + return torch.nn.Conv2d_forward_before_lora(self, input) + + +def lora_Conv2d_load_state_dict(self, *args, **kwargs): + lora_reset_cached_weight(self) + + return torch.nn.Conv2d_load_state_dict_before_lora(self, *args, **kwargs) + + +def lora_MultiheadAttention_forward(self, *args, **kwargs): + lora_apply_weights(self) + + return torch.nn.MultiheadAttention_forward_before_lora(self, *args, **kwargs) + + +def lora_MultiheadAttention_load_state_dict(self, *args, **kwargs): + lora_reset_cached_weight(self) + + return torch.nn.MultiheadAttention_load_state_dict_before_lora(self, *args, **kwargs) + + +def list_available_loras(): + available_loras.clear() + available_lora_aliases.clear() + forbidden_lora_aliases.clear() + available_lora_hash_lookup.clear() + forbidden_lora_aliases.update({"none": 1, "Addams": 1}) + + os.makedirs(shared.cmd_opts.lora_dir, exist_ok=True) + + candidates = list(shared.walk_files(shared.cmd_opts.lora_dir, allowed_extensions=[".pt", ".ckpt", ".safetensors"])) + for filename in sorted(candidates, key=str.lower): + if os.path.isdir(filename): + continue + + name = os.path.splitext(os.path.basename(filename))[0] + entry = LoraOnDisk(name, filename) + + available_loras[name] = entry + + if entry.alias in available_lora_aliases: + forbidden_lora_aliases[entry.alias.lower()] = 1 + + available_lora_aliases[name] = entry + available_lora_aliases[entry.alias] = entry + + +re_lora_name = re.compile(r"(.*)\s*\([0-9a-fA-F]+\)") + + +def infotext_pasted(infotext, params): + if "AddNet Module 1" in [x[1] for x in scripts.scripts_txt2img.infotext_fields]: + return # if the other extension is active, it will handle those fields, no need to do anything + + added = [] + + for k in params: + if not k.startswith("AddNet Model "): + continue + + num = k[13:] + + if params.get("AddNet Module " + num) != "LoRA": + continue + + name = params.get("AddNet Model " + num) + if name is None: + continue + + m = re_lora_name.match(name) + if m: + name = m.group(1) + + multiplier = params.get("AddNet Weight A " + num, "1.0") + + added.append(f"") + + if added: + params["Prompt"] += "\n" + "".join(added) + + +available_loras = {} +available_lora_aliases = {} +available_lora_hash_lookup = {} +forbidden_lora_aliases = {} +loaded_loras = [] + +list_available_loras() diff --git a/extensions-builtin/Lora/preload.py b/extensions-builtin/Lora/preload.py new file mode 100644 index 0000000000000000000000000000000000000000..c47d7ef4e24893953e51b8dfb5a7ccf88c574546 --- /dev/null +++ b/extensions-builtin/Lora/preload.py @@ -0,0 +1,6 @@ +import os +from modules import paths + + +def preload(parser): + parser.add_argument("--lora-dir", type=str, help="Path to directory with Lora networks.", default=os.path.join(paths.models_path, 'Lora')) diff --git a/extensions-builtin/Lora/scripts/lora_script.py b/extensions-builtin/Lora/scripts/lora_script.py new file mode 100644 index 0000000000000000000000000000000000000000..891c1bb164d208654e565b1ba211e25c89078c31 --- /dev/null +++ b/extensions-builtin/Lora/scripts/lora_script.py @@ -0,0 +1,116 @@ +import re + +import torch +import gradio as gr +from fastapi import FastAPI + +import lora +import extra_networks_lora +import ui_extra_networks_lora +from modules import script_callbacks, ui_extra_networks, extra_networks, shared + +def unload(): + torch.nn.Linear.forward = torch.nn.Linear_forward_before_lora + torch.nn.Linear._load_from_state_dict = torch.nn.Linear_load_state_dict_before_lora + torch.nn.Conv2d.forward = torch.nn.Conv2d_forward_before_lora + torch.nn.Conv2d._load_from_state_dict = torch.nn.Conv2d_load_state_dict_before_lora + torch.nn.MultiheadAttention.forward = torch.nn.MultiheadAttention_forward_before_lora + torch.nn.MultiheadAttention._load_from_state_dict = torch.nn.MultiheadAttention_load_state_dict_before_lora + + +def before_ui(): + ui_extra_networks.register_page(ui_extra_networks_lora.ExtraNetworksPageLora()) + extra_networks.register_extra_network(extra_networks_lora.ExtraNetworkLora()) + + +if not hasattr(torch.nn, 'Linear_forward_before_lora'): + torch.nn.Linear_forward_before_lora = torch.nn.Linear.forward + +if not hasattr(torch.nn, 'Linear_load_state_dict_before_lora'): + torch.nn.Linear_load_state_dict_before_lora = torch.nn.Linear._load_from_state_dict + +if not hasattr(torch.nn, 'Conv2d_forward_before_lora'): + torch.nn.Conv2d_forward_before_lora = torch.nn.Conv2d.forward + +if not hasattr(torch.nn, 'Conv2d_load_state_dict_before_lora'): + torch.nn.Conv2d_load_state_dict_before_lora = torch.nn.Conv2d._load_from_state_dict + +if not hasattr(torch.nn, 'MultiheadAttention_forward_before_lora'): + torch.nn.MultiheadAttention_forward_before_lora = torch.nn.MultiheadAttention.forward + +if not hasattr(torch.nn, 'MultiheadAttention_load_state_dict_before_lora'): + torch.nn.MultiheadAttention_load_state_dict_before_lora = torch.nn.MultiheadAttention._load_from_state_dict + +torch.nn.Linear.forward = lora.lora_Linear_forward +torch.nn.Linear._load_from_state_dict = lora.lora_Linear_load_state_dict +torch.nn.Conv2d.forward = lora.lora_Conv2d_forward +torch.nn.Conv2d._load_from_state_dict = lora.lora_Conv2d_load_state_dict +torch.nn.MultiheadAttention.forward = lora.lora_MultiheadAttention_forward +torch.nn.MultiheadAttention._load_from_state_dict = lora.lora_MultiheadAttention_load_state_dict + +script_callbacks.on_model_loaded(lora.assign_lora_names_to_compvis_modules) +script_callbacks.on_script_unloaded(unload) +script_callbacks.on_before_ui(before_ui) +script_callbacks.on_infotext_pasted(lora.infotext_pasted) + + +shared.options_templates.update(shared.options_section(('extra_networks', "Extra Networks"), { + "sd_lora": shared.OptionInfo("None", "Add Lora to prompt", gr.Dropdown, lambda: {"choices": ["None", *lora.available_loras]}, refresh=lora.list_available_loras), + "lora_preferred_name": shared.OptionInfo("Alias from file", "When adding to prompt, refer to Lora by", gr.Radio, {"choices": ["Alias from file", "Filename"]}), + "lora_add_hashes_to_infotext": shared.OptionInfo(True, "Add Lora hashes to infotext"), +})) + + +shared.options_templates.update(shared.options_section(('compatibility', "Compatibility"), { + "lora_functional": shared.OptionInfo(False, "Lora: use old method that takes longer when you have multiple Loras active and produces same results as kohya-ss/sd-webui-additional-networks extension"), +})) + + +def create_lora_json(obj: lora.LoraOnDisk): + return { + "name": obj.name, + "alias": obj.alias, + "path": obj.filename, + "metadata": obj.metadata, + } + + +def api_loras(_: gr.Blocks, app: FastAPI): + @app.get("/sdapi/v1/loras") + async def get_loras(): + return [create_lora_json(obj) for obj in lora.available_loras.values()] + + @app.post("/sdapi/v1/refresh-loras") + async def refresh_loras(): + return lora.list_available_loras() + + +script_callbacks.on_app_started(api_loras) + +re_lora = re.compile(""), + "local_preview": f"{path}.{shared.opts.samples_format}", + "metadata": json.dumps(lora_on_disk.metadata, indent=4) if lora_on_disk.metadata else None, + } + + def allowed_directories_for_previews(self): + return [shared.cmd_opts.lora_dir] + diff --git a/extensions-builtin/ScuNET/preload.py b/extensions-builtin/ScuNET/preload.py new file mode 100644 index 0000000000000000000000000000000000000000..4ce82b1d4349b24192b1915d022ed4fda9f31e5c --- /dev/null +++ b/extensions-builtin/ScuNET/preload.py @@ -0,0 +1,6 @@ +import os +from modules import paths + + +def preload(parser): + parser.add_argument("--scunet-models-path", type=str, help="Path to directory with ScuNET model file(s).", default=os.path.join(paths.models_path, 'ScuNET')) diff --git a/extensions-builtin/ScuNET/scripts/scunet_model.py b/extensions-builtin/ScuNET/scripts/scunet_model.py new file mode 100644 index 0000000000000000000000000000000000000000..45d9297b6a194b5c089f58fbddf148fba5a2291e --- /dev/null +++ b/extensions-builtin/ScuNET/scripts/scunet_model.py @@ -0,0 +1,149 @@ +import os.path +import sys +import traceback + +import PIL.Image +import numpy as np +import torch +from tqdm import tqdm + +from basicsr.utils.download_util import load_file_from_url + +import modules.upscaler +from modules import devices, modelloader, script_callbacks +from scunet_model_arch import SCUNet as net +from modules.shared import opts + + +class UpscalerScuNET(modules.upscaler.Upscaler): + def __init__(self, dirname): + self.name = "ScuNET" + self.model_name = "ScuNET GAN" + self.model_name2 = "ScuNET PSNR" + self.model_url = "https://github.com/cszn/KAIR/releases/download/v1.0/scunet_color_real_gan.pth" + self.model_url2 = "https://github.com/cszn/KAIR/releases/download/v1.0/scunet_color_real_psnr.pth" + self.user_path = dirname + super().__init__() + model_paths = self.find_models(ext_filter=[".pth"]) + scalers = [] + add_model2 = True + for file in model_paths: + if "http" in file: + name = self.model_name + else: + name = modelloader.friendly_name(file) + if name == self.model_name2 or file == self.model_url2: + add_model2 = False + try: + scaler_data = modules.upscaler.UpscalerData(name, file, self, 4) + scalers.append(scaler_data) + except Exception: + print(f"Error loading ScuNET model: {file}", file=sys.stderr) + print(traceback.format_exc(), file=sys.stderr) + if add_model2: + scaler_data2 = modules.upscaler.UpscalerData(self.model_name2, self.model_url2, self) + scalers.append(scaler_data2) + self.scalers = scalers + + @staticmethod + @torch.no_grad() + def tiled_inference(img, model): + # test the image tile by tile + h, w = img.shape[2:] + tile = opts.SCUNET_tile + tile_overlap = opts.SCUNET_tile_overlap + if tile == 0: + return model(img) + + device = devices.get_device_for('scunet') + assert tile % 8 == 0, "tile size should be a multiple of window_size" + sf = 1 + + stride = tile - tile_overlap + h_idx_list = list(range(0, h - tile, stride)) + [h - tile] + w_idx_list = list(range(0, w - tile, stride)) + [w - tile] + E = torch.zeros(1, 3, h * sf, w * sf, dtype=img.dtype, device=device) + W = torch.zeros_like(E, dtype=devices.dtype, device=device) + + with tqdm(total=len(h_idx_list) * len(w_idx_list), desc="ScuNET tiles") as pbar: + for h_idx in h_idx_list: + + for w_idx in w_idx_list: + + in_patch = img[..., h_idx: h_idx + tile, w_idx: w_idx + tile] + + out_patch = model(in_patch) + out_patch_mask = torch.ones_like(out_patch) + + E[ + ..., h_idx * sf: (h_idx + tile) * sf, w_idx * sf: (w_idx + tile) * sf + ].add_(out_patch) + W[ + ..., h_idx * sf: (h_idx + tile) * sf, w_idx * sf: (w_idx + tile) * sf + ].add_(out_patch_mask) + pbar.update(1) + output = E.div_(W) + + return output + + def do_upscale(self, img: PIL.Image.Image, selected_file): + + torch.cuda.empty_cache() + + model = self.load_model(selected_file) + if model is None: + print(f"ScuNET: Unable to load model from {selected_file}", file=sys.stderr) + return img + + device = devices.get_device_for('scunet') + tile = opts.SCUNET_tile + h, w = img.height, img.width + np_img = np.array(img) + np_img = np_img[:, :, ::-1] # RGB to BGR + np_img = np_img.transpose((2, 0, 1)) / 255 # HWC to CHW + torch_img = torch.from_numpy(np_img).float().unsqueeze(0).to(device) # type: ignore + + if tile > h or tile > w: + _img = torch.zeros(1, 3, max(h, tile), max(w, tile), dtype=torch_img.dtype, device=torch_img.device) + _img[:, :, :h, :w] = torch_img # pad image + torch_img = _img + + torch_output = self.tiled_inference(torch_img, model).squeeze(0) + torch_output = torch_output[:, :h * 1, :w * 1] # remove padding, if any + np_output: np.ndarray = torch_output.float().cpu().clamp_(0, 1).numpy() + del torch_img, torch_output + torch.cuda.empty_cache() + + output = np_output.transpose((1, 2, 0)) # CHW to HWC + output = output[:, :, ::-1] # BGR to RGB + return PIL.Image.fromarray((output * 255).astype(np.uint8)) + + def load_model(self, path: str): + device = devices.get_device_for('scunet') + if "http" in path: + filename = load_file_from_url(url=self.model_url, model_dir=self.model_download_path, file_name="%s.pth" % self.name, progress=True) + else: + filename = path + if not os.path.exists(os.path.join(self.model_path, filename)) or filename is None: + print(f"ScuNET: Unable to load model from {filename}", file=sys.stderr) + return None + + model = net(in_nc=3, config=[4, 4, 4, 4, 4, 4, 4], dim=64) + model.load_state_dict(torch.load(filename), strict=True) + model.eval() + for _, v in model.named_parameters(): + v.requires_grad = False + model = model.to(device) + + return model + + +def on_ui_settings(): + import gradio as gr + from modules import shared + + shared.opts.add_option("SCUNET_tile", shared.OptionInfo(256, "Tile size for SCUNET upscalers.", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}, section=('upscaling', "Upscaling")).info("0 = no tiling")) + shared.opts.add_option("SCUNET_tile_overlap", shared.OptionInfo(8, "Tile overlap for SCUNET upscalers.", gr.Slider, {"minimum": 0, "maximum": 64, "step": 1}, section=('upscaling', "Upscaling")).info("Low values = visible seam")) + + +script_callbacks.on_ui_settings(on_ui_settings) diff --git a/extensions-builtin/ScuNET/scunet_model_arch.py b/extensions-builtin/ScuNET/scunet_model_arch.py new file mode 100644 index 0000000000000000000000000000000000000000..b51a880629baa492ffcbebe682bcf101f06699a6 --- /dev/null +++ b/extensions-builtin/ScuNET/scunet_model_arch.py @@ -0,0 +1,268 @@ +# -*- coding: utf-8 -*- +import numpy as np +import torch +import torch.nn as nn +from einops import rearrange +from einops.layers.torch import Rearrange +from timm.models.layers import trunc_normal_, DropPath + + +class WMSA(nn.Module): + """ Self-attention module in Swin Transformer + """ + + def __init__(self, input_dim, output_dim, head_dim, window_size, type): + super(WMSA, self).__init__() + self.input_dim = input_dim + self.output_dim = output_dim + self.head_dim = head_dim + self.scale = self.head_dim ** -0.5 + self.n_heads = input_dim // head_dim + self.window_size = window_size + self.type = type + self.embedding_layer = nn.Linear(self.input_dim, 3 * self.input_dim, bias=True) + + self.relative_position_params = nn.Parameter( + torch.zeros((2 * window_size - 1) * (2 * window_size - 1), self.n_heads)) + + self.linear = nn.Linear(self.input_dim, self.output_dim) + + trunc_normal_(self.relative_position_params, std=.02) + self.relative_position_params = torch.nn.Parameter( + self.relative_position_params.view(2 * window_size - 1, 2 * window_size - 1, self.n_heads).transpose(1, + 2).transpose( + 0, 1)) + + def generate_mask(self, h, w, p, shift): + """ generating the mask of SW-MSA + Args: + shift: shift parameters in CyclicShift. + Returns: + attn_mask: should be (1 1 w p p), + """ + # supporting square. + attn_mask = torch.zeros(h, w, p, p, p, p, dtype=torch.bool, device=self.relative_position_params.device) + if self.type == 'W': + return attn_mask + + s = p - shift + attn_mask[-1, :, :s, :, s:, :] = True + attn_mask[-1, :, s:, :, :s, :] = True + attn_mask[:, -1, :, :s, :, s:] = True + attn_mask[:, -1, :, s:, :, :s] = True + attn_mask = rearrange(attn_mask, 'w1 w2 p1 p2 p3 p4 -> 1 1 (w1 w2) (p1 p2) (p3 p4)') + return attn_mask + + def forward(self, x): + """ Forward pass of Window Multi-head Self-attention module. + Args: + x: input tensor with shape of [b h w c]; + attn_mask: attention mask, fill -inf where the value is True; + Returns: + output: tensor shape [b h w c] + """ + if self.type != 'W': + x = torch.roll(x, shifts=(-(self.window_size // 2), -(self.window_size // 2)), dims=(1, 2)) + + x = rearrange(x, 'b (w1 p1) (w2 p2) c -> b w1 w2 p1 p2 c', p1=self.window_size, p2=self.window_size) + h_windows = x.size(1) + w_windows = x.size(2) + # square validation + # assert h_windows == w_windows + + x = rearrange(x, 'b w1 w2 p1 p2 c -> b (w1 w2) (p1 p2) c', p1=self.window_size, p2=self.window_size) + qkv = self.embedding_layer(x) + q, k, v = rearrange(qkv, 'b nw np (threeh c) -> threeh b nw np c', c=self.head_dim).chunk(3, dim=0) + sim = torch.einsum('hbwpc,hbwqc->hbwpq', q, k) * self.scale + # Adding learnable relative embedding + sim = sim + rearrange(self.relative_embedding(), 'h p q -> h 1 1 p q') + # Using Attn Mask to distinguish different subwindows. + if self.type != 'W': + attn_mask = self.generate_mask(h_windows, w_windows, self.window_size, shift=self.window_size // 2) + sim = sim.masked_fill_(attn_mask, float("-inf")) + + probs = nn.functional.softmax(sim, dim=-1) + output = torch.einsum('hbwij,hbwjc->hbwic', probs, v) + output = rearrange(output, 'h b w p c -> b w p (h c)') + output = self.linear(output) + output = rearrange(output, 'b (w1 w2) (p1 p2) c -> b (w1 p1) (w2 p2) c', w1=h_windows, p1=self.window_size) + + if self.type != 'W': + output = torch.roll(output, shifts=(self.window_size // 2, self.window_size // 2), dims=(1, 2)) + + return output + + def relative_embedding(self): + cord = torch.tensor(np.array([[i, j] for i in range(self.window_size) for j in range(self.window_size)])) + relation = cord[:, None, :] - cord[None, :, :] + self.window_size - 1 + # negative is allowed + return self.relative_position_params[:, relation[:, :, 0].long(), relation[:, :, 1].long()] + + +class Block(nn.Module): + def __init__(self, input_dim, output_dim, head_dim, window_size, drop_path, type='W', input_resolution=None): + """ SwinTransformer Block + """ + super(Block, self).__init__() + self.input_dim = input_dim + self.output_dim = output_dim + assert type in ['W', 'SW'] + self.type = type + if input_resolution <= window_size: + self.type = 'W' + + self.ln1 = nn.LayerNorm(input_dim) + self.msa = WMSA(input_dim, input_dim, head_dim, window_size, self.type) + self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity() + self.ln2 = nn.LayerNorm(input_dim) + self.mlp = nn.Sequential( + nn.Linear(input_dim, 4 * input_dim), + nn.GELU(), + nn.Linear(4 * input_dim, output_dim), + ) + + def forward(self, x): + x = x + self.drop_path(self.msa(self.ln1(x))) + x = x + self.drop_path(self.mlp(self.ln2(x))) + return x + + +class ConvTransBlock(nn.Module): + def __init__(self, conv_dim, trans_dim, head_dim, window_size, drop_path, type='W', input_resolution=None): + """ SwinTransformer and Conv Block + """ + super(ConvTransBlock, self).__init__() + self.conv_dim = conv_dim + self.trans_dim = trans_dim + self.head_dim = head_dim + self.window_size = window_size + self.drop_path = drop_path + self.type = type + self.input_resolution = input_resolution + + assert self.type in ['W', 'SW'] + if self.input_resolution <= self.window_size: + self.type = 'W' + + self.trans_block = Block(self.trans_dim, self.trans_dim, self.head_dim, self.window_size, self.drop_path, + self.type, self.input_resolution) + self.conv1_1 = nn.Conv2d(self.conv_dim + self.trans_dim, self.conv_dim + self.trans_dim, 1, 1, 0, bias=True) + self.conv1_2 = nn.Conv2d(self.conv_dim + self.trans_dim, self.conv_dim + self.trans_dim, 1, 1, 0, bias=True) + + self.conv_block = nn.Sequential( + nn.Conv2d(self.conv_dim, self.conv_dim, 3, 1, 1, bias=False), + nn.ReLU(True), + nn.Conv2d(self.conv_dim, self.conv_dim, 3, 1, 1, bias=False) + ) + + def forward(self, x): + conv_x, trans_x = torch.split(self.conv1_1(x), (self.conv_dim, self.trans_dim), dim=1) + conv_x = self.conv_block(conv_x) + conv_x + trans_x = Rearrange('b c h w -> b h w c')(trans_x) + trans_x = self.trans_block(trans_x) + trans_x = Rearrange('b h w c -> b c h w')(trans_x) + res = self.conv1_2(torch.cat((conv_x, trans_x), dim=1)) + x = x + res + + return x + + +class SCUNet(nn.Module): + # def __init__(self, in_nc=3, config=[2, 2, 2, 2, 2, 2, 2], dim=64, drop_path_rate=0.0, input_resolution=256): + def __init__(self, in_nc=3, config=None, dim=64, drop_path_rate=0.0, input_resolution=256): + super(SCUNet, self).__init__() + if config is None: + config = [2, 2, 2, 2, 2, 2, 2] + self.config = config + self.dim = dim + self.head_dim = 32 + self.window_size = 8 + + # drop path rate for each layer + dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(config))] + + self.m_head = [nn.Conv2d(in_nc, dim, 3, 1, 1, bias=False)] + + begin = 0 + self.m_down1 = [ConvTransBlock(dim // 2, dim // 2, self.head_dim, self.window_size, dpr[i + begin], + 'W' if not i % 2 else 'SW', input_resolution) + for i in range(config[0])] + \ + [nn.Conv2d(dim, 2 * dim, 2, 2, 0, bias=False)] + + begin += config[0] + self.m_down2 = [ConvTransBlock(dim, dim, self.head_dim, self.window_size, dpr[i + begin], + 'W' if not i % 2 else 'SW', input_resolution // 2) + for i in range(config[1])] + \ + [nn.Conv2d(2 * dim, 4 * dim, 2, 2, 0, bias=False)] + + begin += config[1] + self.m_down3 = [ConvTransBlock(2 * dim, 2 * dim, self.head_dim, self.window_size, dpr[i + begin], + 'W' if not i % 2 else 'SW', input_resolution // 4) + for i in range(config[2])] + \ + [nn.Conv2d(4 * dim, 8 * dim, 2, 2, 0, bias=False)] + + begin += config[2] + self.m_body = [ConvTransBlock(4 * dim, 4 * dim, self.head_dim, self.window_size, dpr[i + begin], + 'W' if not i % 2 else 'SW', input_resolution // 8) + for i in range(config[3])] + + begin += config[3] + self.m_up3 = [nn.ConvTranspose2d(8 * dim, 4 * dim, 2, 2, 0, bias=False), ] + \ + [ConvTransBlock(2 * dim, 2 * dim, self.head_dim, self.window_size, dpr[i + begin], + 'W' if not i % 2 else 'SW', input_resolution // 4) + for i in range(config[4])] + + begin += config[4] + self.m_up2 = [nn.ConvTranspose2d(4 * dim, 2 * dim, 2, 2, 0, bias=False), ] + \ + [ConvTransBlock(dim, dim, self.head_dim, self.window_size, dpr[i + begin], + 'W' if not i % 2 else 'SW', input_resolution // 2) + for i in range(config[5])] + + begin += config[5] + self.m_up1 = [nn.ConvTranspose2d(2 * dim, dim, 2, 2, 0, bias=False), ] + \ + [ConvTransBlock(dim // 2, dim // 2, self.head_dim, self.window_size, dpr[i + begin], + 'W' if not i % 2 else 'SW', input_resolution) + for i in range(config[6])] + + self.m_tail = [nn.Conv2d(dim, in_nc, 3, 1, 1, bias=False)] + + self.m_head = nn.Sequential(*self.m_head) + self.m_down1 = nn.Sequential(*self.m_down1) + self.m_down2 = nn.Sequential(*self.m_down2) + self.m_down3 = nn.Sequential(*self.m_down3) + self.m_body = nn.Sequential(*self.m_body) + self.m_up3 = nn.Sequential(*self.m_up3) + self.m_up2 = nn.Sequential(*self.m_up2) + self.m_up1 = nn.Sequential(*self.m_up1) + self.m_tail = nn.Sequential(*self.m_tail) + # self.apply(self._init_weights) + + def forward(self, x0): + + h, w = x0.size()[-2:] + paddingBottom = int(np.ceil(h / 64) * 64 - h) + paddingRight = int(np.ceil(w / 64) * 64 - w) + x0 = nn.ReplicationPad2d((0, paddingRight, 0, paddingBottom))(x0) + + x1 = self.m_head(x0) + x2 = self.m_down1(x1) + x3 = self.m_down2(x2) + x4 = self.m_down3(x3) + x = self.m_body(x4) + x = self.m_up3(x + x4) + x = self.m_up2(x + x3) + x = self.m_up1(x + x2) + x = self.m_tail(x + x1) + + x = x[..., :h, :w] + + return x + + def _init_weights(self, m): + if isinstance(m, nn.Linear): + trunc_normal_(m.weight, std=.02) + if m.bias is not None: + nn.init.constant_(m.bias, 0) + elif isinstance(m, nn.LayerNorm): + nn.init.constant_(m.bias, 0) + nn.init.constant_(m.weight, 1.0) diff --git a/extensions-builtin/SwinIR/preload.py b/extensions-builtin/SwinIR/preload.py new file mode 100644 index 0000000000000000000000000000000000000000..e912c6402bc80faa797cf2e95183101fb9a10286 --- /dev/null +++ b/extensions-builtin/SwinIR/preload.py @@ -0,0 +1,6 @@ +import os +from modules import paths + + +def preload(parser): + parser.add_argument("--swinir-models-path", type=str, help="Path to directory with SwinIR model file(s).", default=os.path.join(paths.models_path, 'SwinIR')) diff --git a/extensions-builtin/SwinIR/scripts/swinir_model.py b/extensions-builtin/SwinIR/scripts/swinir_model.py new file mode 100644 index 0000000000000000000000000000000000000000..1c7bf325e999396aa88ff5fd56311ac9e1357773 --- /dev/null +++ b/extensions-builtin/SwinIR/scripts/swinir_model.py @@ -0,0 +1,177 @@ +import os + +import numpy as np +import torch +from PIL import Image +from basicsr.utils.download_util import load_file_from_url +from tqdm import tqdm + +from modules import modelloader, devices, script_callbacks, shared +from modules.shared import opts, state +from swinir_model_arch import SwinIR as net +from swinir_model_arch_v2 import Swin2SR as net2 +from modules.upscaler import Upscaler, UpscalerData + + +device_swinir = devices.get_device_for('swinir') + + +class UpscalerSwinIR(Upscaler): + def __init__(self, dirname): + self.name = "SwinIR" + self.model_url = "https://github.com/JingyunLiang/SwinIR/releases/download/v0.0" \ + "/003_realSR_BSRGAN_DFOWMFC_s64w8_SwinIR" \ + "-L_x4_GAN.pth " + self.model_name = "SwinIR 4x" + self.user_path = dirname + super().__init__() + scalers = [] + model_files = self.find_models(ext_filter=[".pt", ".pth"]) + for model in model_files: + if "http" in model: + name = self.model_name + else: + name = modelloader.friendly_name(model) + model_data = UpscalerData(name, model, self) + scalers.append(model_data) + self.scalers = scalers + + def do_upscale(self, img, model_file): + model = self.load_model(model_file) + if model is None: + return img + model = model.to(device_swinir, dtype=devices.dtype) + img = upscale(img, model) + try: + torch.cuda.empty_cache() + except Exception: + pass + return img + + def load_model(self, path, scale=4): + if "http" in path: + dl_name = "%s%s" % (self.model_name.replace(" ", "_"), ".pth") + filename = load_file_from_url(url=path, model_dir=self.model_download_path, file_name=dl_name, progress=True) + else: + filename = path + if filename is None or not os.path.exists(filename): + return None + if filename.endswith(".v2.pth"): + model = net2( + upscale=scale, + in_chans=3, + img_size=64, + window_size=8, + img_range=1.0, + depths=[6, 6, 6, 6, 6, 6], + embed_dim=180, + num_heads=[6, 6, 6, 6, 6, 6], + mlp_ratio=2, + upsampler="nearest+conv", + resi_connection="1conv", + ) + params = None + else: + model = net( + upscale=scale, + in_chans=3, + img_size=64, + window_size=8, + img_range=1.0, + depths=[6, 6, 6, 6, 6, 6, 6, 6, 6], + embed_dim=240, + num_heads=[8, 8, 8, 8, 8, 8, 8, 8, 8], + mlp_ratio=2, + upsampler="nearest+conv", + resi_connection="3conv", + ) + params = "params_ema" + + pretrained_model = torch.load(filename) + if params is not None: + model.load_state_dict(pretrained_model[params], strict=True) + else: + model.load_state_dict(pretrained_model, strict=True) + return model + + +def upscale( + img, + model, + tile=None, + tile_overlap=None, + window_size=8, + scale=4, +): + tile = tile or opts.SWIN_tile + tile_overlap = tile_overlap or opts.SWIN_tile_overlap + + + img = np.array(img) + img = img[:, :, ::-1] + img = np.moveaxis(img, 2, 0) / 255 + img = torch.from_numpy(img).float() + img = img.unsqueeze(0).to(device_swinir, dtype=devices.dtype) + with torch.no_grad(), devices.autocast(): + _, _, h_old, w_old = img.size() + h_pad = (h_old // window_size + 1) * window_size - h_old + w_pad = (w_old // window_size + 1) * window_size - w_old + img = torch.cat([img, torch.flip(img, [2])], 2)[:, :, : h_old + h_pad, :] + img = torch.cat([img, torch.flip(img, [3])], 3)[:, :, :, : w_old + w_pad] + output = inference(img, model, tile, tile_overlap, window_size, scale) + output = output[..., : h_old * scale, : w_old * scale] + output = output.data.squeeze().float().cpu().clamp_(0, 1).numpy() + if output.ndim == 3: + output = np.transpose( + output[[2, 1, 0], :, :], (1, 2, 0) + ) # CHW-RGB to HCW-BGR + output = (output * 255.0).round().astype(np.uint8) # float32 to uint8 + return Image.fromarray(output, "RGB") + + +def inference(img, model, tile, tile_overlap, window_size, scale): + # test the image tile by tile + b, c, h, w = img.size() + tile = min(tile, h, w) + assert tile % window_size == 0, "tile size should be a multiple of window_size" + sf = scale + + stride = tile - tile_overlap + h_idx_list = list(range(0, h - tile, stride)) + [h - tile] + w_idx_list = list(range(0, w - tile, stride)) + [w - tile] + E = torch.zeros(b, c, h * sf, w * sf, dtype=devices.dtype, device=device_swinir).type_as(img) + W = torch.zeros_like(E, dtype=devices.dtype, device=device_swinir) + + with tqdm(total=len(h_idx_list) * len(w_idx_list), desc="SwinIR tiles") as pbar: + for h_idx in h_idx_list: + if state.interrupted or state.skipped: + break + + for w_idx in w_idx_list: + if state.interrupted or state.skipped: + break + + in_patch = img[..., h_idx: h_idx + tile, w_idx: w_idx + tile] + out_patch = model(in_patch) + out_patch_mask = torch.ones_like(out_patch) + + E[ + ..., h_idx * sf: (h_idx + tile) * sf, w_idx * sf: (w_idx + tile) * sf + ].add_(out_patch) + W[ + ..., h_idx * sf: (h_idx + tile) * sf, w_idx * sf: (w_idx + tile) * sf + ].add_(out_patch_mask) + pbar.update(1) + output = E.div_(W) + + return output + + +def on_ui_settings(): + import gradio as gr + + shared.opts.add_option("SWIN_tile", shared.OptionInfo(192, "Tile size for all SwinIR.", gr.Slider, {"minimum": 16, "maximum": 512, "step": 16}, section=('upscaling', "Upscaling"))) + shared.opts.add_option("SWIN_tile_overlap", shared.OptionInfo(8, "Tile overlap, in pixels for SwinIR. Low values = visible seam.", gr.Slider, {"minimum": 0, "maximum": 48, "step": 1}, section=('upscaling', "Upscaling"))) + + +script_callbacks.on_ui_settings(on_ui_settings) diff --git a/extensions-builtin/SwinIR/swinir_model_arch.py b/extensions-builtin/SwinIR/swinir_model_arch.py new file mode 100644 index 0000000000000000000000000000000000000000..93b9327473a6e77c3a3dc6a7743e932c9083a996 --- /dev/null +++ b/extensions-builtin/SwinIR/swinir_model_arch.py @@ -0,0 +1,867 @@ +# ----------------------------------------------------------------------------------- +# SwinIR: Image Restoration Using Swin Transformer, https://arxiv.org/abs/2108.10257 +# Originally Written by Ze Liu, Modified by Jingyun Liang. +# ----------------------------------------------------------------------------------- + +import math +import torch +import torch.nn as nn +import torch.nn.functional as F +import torch.utils.checkpoint as checkpoint +from timm.models.layers import DropPath, to_2tuple, trunc_normal_ + + +class Mlp(nn.Module): + def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.): + super().__init__() + out_features = out_features or in_features + hidden_features = hidden_features or in_features + self.fc1 = nn.Linear(in_features, hidden_features) + self.act = act_layer() + self.fc2 = nn.Linear(hidden_features, out_features) + self.drop = nn.Dropout(drop) + + def forward(self, x): + x = self.fc1(x) + x = self.act(x) + x = self.drop(x) + x = self.fc2(x) + x = self.drop(x) + return x + + +def window_partition(x, window_size): + """ + Args: + x: (B, H, W, C) + window_size (int): window size + + Returns: + windows: (num_windows*B, window_size, window_size, C) + """ + B, H, W, C = x.shape + x = x.view(B, H // window_size, window_size, W // window_size, window_size, C) + windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C) + return windows + + +def window_reverse(windows, window_size, H, W): + """ + Args: + windows: (num_windows*B, window_size, window_size, C) + window_size (int): Window size + H (int): Height of image + W (int): Width of image + + Returns: + x: (B, H, W, C) + """ + B = int(windows.shape[0] / (H * W / window_size / window_size)) + x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1) + x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1) + return x + + +class WindowAttention(nn.Module): + r""" Window based multi-head self attention (W-MSA) module with relative position bias. + It supports both of shifted and non-shifted window. + + Args: + dim (int): Number of input channels. + window_size (tuple[int]): The height and width of the window. + num_heads (int): Number of attention heads. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set + attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0 + proj_drop (float, optional): Dropout ratio of output. Default: 0.0 + """ + + def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.): + + super().__init__() + self.dim = dim + self.window_size = window_size # Wh, Ww + self.num_heads = num_heads + head_dim = dim // num_heads + self.scale = qk_scale or head_dim ** -0.5 + + # define a parameter table of relative position bias + self.relative_position_bias_table = nn.Parameter( + torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads)) # 2*Wh-1 * 2*Ww-1, nH + + # get pair-wise relative position index for each token inside the window + coords_h = torch.arange(self.window_size[0]) + coords_w = torch.arange(self.window_size[1]) + coords = torch.stack(torch.meshgrid([coords_h, coords_w])) # 2, Wh, Ww + coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww + relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww + relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2 + relative_coords[:, :, 0] += self.window_size[0] - 1 # shift to start from 0 + relative_coords[:, :, 1] += self.window_size[1] - 1 + relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1 + relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww + self.register_buffer("relative_position_index", relative_position_index) + + self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias) + self.attn_drop = nn.Dropout(attn_drop) + self.proj = nn.Linear(dim, dim) + + self.proj_drop = nn.Dropout(proj_drop) + + trunc_normal_(self.relative_position_bias_table, std=.02) + self.softmax = nn.Softmax(dim=-1) + + def forward(self, x, mask=None): + """ + Args: + x: input features with shape of (num_windows*B, N, C) + mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None + """ + B_, N, C = x.shape + qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4) + q, k, v = qkv[0], qkv[1], qkv[2] # make torchscript happy (cannot use tensor as tuple) + + q = q * self.scale + attn = (q @ k.transpose(-2, -1)) + + relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view( + self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1) # Wh*Ww,Wh*Ww,nH + relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous() # nH, Wh*Ww, Wh*Ww + attn = attn + relative_position_bias.unsqueeze(0) + + if mask is not None: + nW = mask.shape[0] + attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0) + attn = attn.view(-1, self.num_heads, N, N) + attn = self.softmax(attn) + else: + attn = self.softmax(attn) + + attn = self.attn_drop(attn) + + x = (attn @ v).transpose(1, 2).reshape(B_, N, C) + x = self.proj(x) + x = self.proj_drop(x) + return x + + def extra_repr(self) -> str: + return f'dim={self.dim}, window_size={self.window_size}, num_heads={self.num_heads}' + + def flops(self, N): + # calculate flops for 1 window with token length of N + flops = 0 + # qkv = self.qkv(x) + flops += N * self.dim * 3 * self.dim + # attn = (q @ k.transpose(-2, -1)) + flops += self.num_heads * N * (self.dim // self.num_heads) * N + # x = (attn @ v) + flops += self.num_heads * N * N * (self.dim // self.num_heads) + # x = self.proj(x) + flops += N * self.dim * self.dim + return flops + + +class SwinTransformerBlock(nn.Module): + r""" Swin Transformer Block. + + Args: + dim (int): Number of input channels. + input_resolution (tuple[int]): Input resolution. + num_heads (int): Number of attention heads. + window_size (int): Window size. + shift_size (int): Shift size for SW-MSA. + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set. + drop (float, optional): Dropout rate. Default: 0.0 + attn_drop (float, optional): Attention dropout rate. Default: 0.0 + drop_path (float, optional): Stochastic depth rate. Default: 0.0 + act_layer (nn.Module, optional): Activation layer. Default: nn.GELU + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + """ + + def __init__(self, dim, input_resolution, num_heads, window_size=7, shift_size=0, + mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., drop_path=0., + act_layer=nn.GELU, norm_layer=nn.LayerNorm): + super().__init__() + self.dim = dim + self.input_resolution = input_resolution + self.num_heads = num_heads + self.window_size = window_size + self.shift_size = shift_size + self.mlp_ratio = mlp_ratio + if min(self.input_resolution) <= self.window_size: + # if window size is larger than input resolution, we don't partition windows + self.shift_size = 0 + self.window_size = min(self.input_resolution) + assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size" + + self.norm1 = norm_layer(dim) + self.attn = WindowAttention( + dim, window_size=to_2tuple(self.window_size), num_heads=num_heads, + qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop) + + self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity() + self.norm2 = norm_layer(dim) + mlp_hidden_dim = int(dim * mlp_ratio) + self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop) + + if self.shift_size > 0: + attn_mask = self.calculate_mask(self.input_resolution) + else: + attn_mask = None + + self.register_buffer("attn_mask", attn_mask) + + def calculate_mask(self, x_size): + # calculate attention mask for SW-MSA + H, W = x_size + img_mask = torch.zeros((1, H, W, 1)) # 1 H W 1 + h_slices = (slice(0, -self.window_size), + slice(-self.window_size, -self.shift_size), + slice(-self.shift_size, None)) + w_slices = (slice(0, -self.window_size), + slice(-self.window_size, -self.shift_size), + slice(-self.shift_size, None)) + cnt = 0 + for h in h_slices: + for w in w_slices: + img_mask[:, h, w, :] = cnt + cnt += 1 + + mask_windows = window_partition(img_mask, self.window_size) # nW, window_size, window_size, 1 + mask_windows = mask_windows.view(-1, self.window_size * self.window_size) + attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2) + attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0)) + + return attn_mask + + def forward(self, x, x_size): + H, W = x_size + B, L, C = x.shape + # assert L == H * W, "input feature has wrong size" + + shortcut = x + x = self.norm1(x) + x = x.view(B, H, W, C) + + # cyclic shift + if self.shift_size > 0: + shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2)) + else: + shifted_x = x + + # partition windows + x_windows = window_partition(shifted_x, self.window_size) # nW*B, window_size, window_size, C + x_windows = x_windows.view(-1, self.window_size * self.window_size, C) # nW*B, window_size*window_size, C + + # W-MSA/SW-MSA (to be compatible for testing on images whose shapes are the multiple of window size + if self.input_resolution == x_size: + attn_windows = self.attn(x_windows, mask=self.attn_mask) # nW*B, window_size*window_size, C + else: + attn_windows = self.attn(x_windows, mask=self.calculate_mask(x_size).to(x.device)) + + # merge windows + attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C) + shifted_x = window_reverse(attn_windows, self.window_size, H, W) # B H' W' C + + # reverse cyclic shift + if self.shift_size > 0: + x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2)) + else: + x = shifted_x + x = x.view(B, H * W, C) + + # FFN + x = shortcut + self.drop_path(x) + x = x + self.drop_path(self.mlp(self.norm2(x))) + + return x + + def extra_repr(self) -> str: + return f"dim={self.dim}, input_resolution={self.input_resolution}, num_heads={self.num_heads}, " \ + f"window_size={self.window_size}, shift_size={self.shift_size}, mlp_ratio={self.mlp_ratio}" + + def flops(self): + flops = 0 + H, W = self.input_resolution + # norm1 + flops += self.dim * H * W + # W-MSA/SW-MSA + nW = H * W / self.window_size / self.window_size + flops += nW * self.attn.flops(self.window_size * self.window_size) + # mlp + flops += 2 * H * W * self.dim * self.dim * self.mlp_ratio + # norm2 + flops += self.dim * H * W + return flops + + +class PatchMerging(nn.Module): + r""" Patch Merging Layer. + + Args: + input_resolution (tuple[int]): Resolution of input feature. + dim (int): Number of input channels. + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + """ + + def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm): + super().__init__() + self.input_resolution = input_resolution + self.dim = dim + self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False) + self.norm = norm_layer(4 * dim) + + def forward(self, x): + """ + x: B, H*W, C + """ + H, W = self.input_resolution + B, L, C = x.shape + assert L == H * W, "input feature has wrong size" + assert H % 2 == 0 and W % 2 == 0, f"x size ({H}*{W}) are not even." + + x = x.view(B, H, W, C) + + x0 = x[:, 0::2, 0::2, :] # B H/2 W/2 C + x1 = x[:, 1::2, 0::2, :] # B H/2 W/2 C + x2 = x[:, 0::2, 1::2, :] # B H/2 W/2 C + x3 = x[:, 1::2, 1::2, :] # B H/2 W/2 C + x = torch.cat([x0, x1, x2, x3], -1) # B H/2 W/2 4*C + x = x.view(B, -1, 4 * C) # B H/2*W/2 4*C + + x = self.norm(x) + x = self.reduction(x) + + return x + + def extra_repr(self) -> str: + return f"input_resolution={self.input_resolution}, dim={self.dim}" + + def flops(self): + H, W = self.input_resolution + flops = H * W * self.dim + flops += (H // 2) * (W // 2) * 4 * self.dim * 2 * self.dim + return flops + + +class BasicLayer(nn.Module): + """ A basic Swin Transformer layer for one stage. + + Args: + dim (int): Number of input channels. + input_resolution (tuple[int]): Input resolution. + depth (int): Number of blocks. + num_heads (int): Number of attention heads. + window_size (int): Local window size. + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set. + drop (float, optional): Dropout rate. Default: 0.0 + attn_drop (float, optional): Attention dropout rate. Default: 0.0 + drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0 + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None + use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False. + """ + + def __init__(self, dim, input_resolution, depth, num_heads, window_size, + mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., + drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False): + + super().__init__() + self.dim = dim + self.input_resolution = input_resolution + self.depth = depth + self.use_checkpoint = use_checkpoint + + # build blocks + self.blocks = nn.ModuleList([ + SwinTransformerBlock(dim=dim, input_resolution=input_resolution, + num_heads=num_heads, window_size=window_size, + shift_size=0 if (i % 2 == 0) else window_size // 2, + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, qk_scale=qk_scale, + drop=drop, attn_drop=attn_drop, + drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path, + norm_layer=norm_layer) + for i in range(depth)]) + + # patch merging layer + if downsample is not None: + self.downsample = downsample(input_resolution, dim=dim, norm_layer=norm_layer) + else: + self.downsample = None + + def forward(self, x, x_size): + for blk in self.blocks: + if self.use_checkpoint: + x = checkpoint.checkpoint(blk, x, x_size) + else: + x = blk(x, x_size) + if self.downsample is not None: + x = self.downsample(x) + return x + + def extra_repr(self) -> str: + return f"dim={self.dim}, input_resolution={self.input_resolution}, depth={self.depth}" + + def flops(self): + flops = 0 + for blk in self.blocks: + flops += blk.flops() + if self.downsample is not None: + flops += self.downsample.flops() + return flops + + +class RSTB(nn.Module): + """Residual Swin Transformer Block (RSTB). + + Args: + dim (int): Number of input channels. + input_resolution (tuple[int]): Input resolution. + depth (int): Number of blocks. + num_heads (int): Number of attention heads. + window_size (int): Local window size. + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set. + drop (float, optional): Dropout rate. Default: 0.0 + attn_drop (float, optional): Attention dropout rate. Default: 0.0 + drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0 + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None + use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False. + img_size: Input image size. + patch_size: Patch size. + resi_connection: The convolutional block before residual connection. + """ + + def __init__(self, dim, input_resolution, depth, num_heads, window_size, + mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., + drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False, + img_size=224, patch_size=4, resi_connection='1conv'): + super(RSTB, self).__init__() + + self.dim = dim + self.input_resolution = input_resolution + + self.residual_group = BasicLayer(dim=dim, + input_resolution=input_resolution, + depth=depth, + num_heads=num_heads, + window_size=window_size, + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, qk_scale=qk_scale, + drop=drop, attn_drop=attn_drop, + drop_path=drop_path, + norm_layer=norm_layer, + downsample=downsample, + use_checkpoint=use_checkpoint) + + if resi_connection == '1conv': + self.conv = nn.Conv2d(dim, dim, 3, 1, 1) + elif resi_connection == '3conv': + # to save parameters and memory + self.conv = nn.Sequential(nn.Conv2d(dim, dim // 4, 3, 1, 1), nn.LeakyReLU(negative_slope=0.2, inplace=True), + nn.Conv2d(dim // 4, dim // 4, 1, 1, 0), + nn.LeakyReLU(negative_slope=0.2, inplace=True), + nn.Conv2d(dim // 4, dim, 3, 1, 1)) + + self.patch_embed = PatchEmbed( + img_size=img_size, patch_size=patch_size, in_chans=0, embed_dim=dim, + norm_layer=None) + + self.patch_unembed = PatchUnEmbed( + img_size=img_size, patch_size=patch_size, in_chans=0, embed_dim=dim, + norm_layer=None) + + def forward(self, x, x_size): + return self.patch_embed(self.conv(self.patch_unembed(self.residual_group(x, x_size), x_size))) + x + + def flops(self): + flops = 0 + flops += self.residual_group.flops() + H, W = self.input_resolution + flops += H * W * self.dim * self.dim * 9 + flops += self.patch_embed.flops() + flops += self.patch_unembed.flops() + + return flops + + +class PatchEmbed(nn.Module): + r""" Image to Patch Embedding + + Args: + img_size (int): Image size. Default: 224. + patch_size (int): Patch token size. Default: 4. + in_chans (int): Number of input image channels. Default: 3. + embed_dim (int): Number of linear projection output channels. Default: 96. + norm_layer (nn.Module, optional): Normalization layer. Default: None + """ + + def __init__(self, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None): + super().__init__() + img_size = to_2tuple(img_size) + patch_size = to_2tuple(patch_size) + patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]] + self.img_size = img_size + self.patch_size = patch_size + self.patches_resolution = patches_resolution + self.num_patches = patches_resolution[0] * patches_resolution[1] + + self.in_chans = in_chans + self.embed_dim = embed_dim + + if norm_layer is not None: + self.norm = norm_layer(embed_dim) + else: + self.norm = None + + def forward(self, x): + x = x.flatten(2).transpose(1, 2) # B Ph*Pw C + if self.norm is not None: + x = self.norm(x) + return x + + def flops(self): + flops = 0 + H, W = self.img_size + if self.norm is not None: + flops += H * W * self.embed_dim + return flops + + +class PatchUnEmbed(nn.Module): + r""" Image to Patch Unembedding + + Args: + img_size (int): Image size. Default: 224. + patch_size (int): Patch token size. Default: 4. + in_chans (int): Number of input image channels. Default: 3. + embed_dim (int): Number of linear projection output channels. Default: 96. + norm_layer (nn.Module, optional): Normalization layer. Default: None + """ + + def __init__(self, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None): + super().__init__() + img_size = to_2tuple(img_size) + patch_size = to_2tuple(patch_size) + patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]] + self.img_size = img_size + self.patch_size = patch_size + self.patches_resolution = patches_resolution + self.num_patches = patches_resolution[0] * patches_resolution[1] + + self.in_chans = in_chans + self.embed_dim = embed_dim + + def forward(self, x, x_size): + B, HW, C = x.shape + x = x.transpose(1, 2).view(B, self.embed_dim, x_size[0], x_size[1]) # B Ph*Pw C + return x + + def flops(self): + flops = 0 + return flops + + +class Upsample(nn.Sequential): + """Upsample module. + + Args: + scale (int): Scale factor. Supported scales: 2^n and 3. + num_feat (int): Channel number of intermediate features. + """ + + def __init__(self, scale, num_feat): + m = [] + if (scale & (scale - 1)) == 0: # scale = 2^n + for _ in range(int(math.log(scale, 2))): + m.append(nn.Conv2d(num_feat, 4 * num_feat, 3, 1, 1)) + m.append(nn.PixelShuffle(2)) + elif scale == 3: + m.append(nn.Conv2d(num_feat, 9 * num_feat, 3, 1, 1)) + m.append(nn.PixelShuffle(3)) + else: + raise ValueError(f'scale {scale} is not supported. ' 'Supported scales: 2^n and 3.') + super(Upsample, self).__init__(*m) + + +class UpsampleOneStep(nn.Sequential): + """UpsampleOneStep module (the difference with Upsample is that it always only has 1conv + 1pixelshuffle) + Used in lightweight SR to save parameters. + + Args: + scale (int): Scale factor. Supported scales: 2^n and 3. + num_feat (int): Channel number of intermediate features. + + """ + + def __init__(self, scale, num_feat, num_out_ch, input_resolution=None): + self.num_feat = num_feat + self.input_resolution = input_resolution + m = [] + m.append(nn.Conv2d(num_feat, (scale ** 2) * num_out_ch, 3, 1, 1)) + m.append(nn.PixelShuffle(scale)) + super(UpsampleOneStep, self).__init__(*m) + + def flops(self): + H, W = self.input_resolution + flops = H * W * self.num_feat * 3 * 9 + return flops + + +class SwinIR(nn.Module): + r""" SwinIR + A PyTorch impl of : `SwinIR: Image Restoration Using Swin Transformer`, based on Swin Transformer. + + Args: + img_size (int | tuple(int)): Input image size. Default 64 + patch_size (int | tuple(int)): Patch size. Default: 1 + in_chans (int): Number of input image channels. Default: 3 + embed_dim (int): Patch embedding dimension. Default: 96 + depths (tuple(int)): Depth of each Swin Transformer layer. + num_heads (tuple(int)): Number of attention heads in different layers. + window_size (int): Window size. Default: 7 + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4 + qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True + qk_scale (float): Override default qk scale of head_dim ** -0.5 if set. Default: None + drop_rate (float): Dropout rate. Default: 0 + attn_drop_rate (float): Attention dropout rate. Default: 0 + drop_path_rate (float): Stochastic depth rate. Default: 0.1 + norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm. + ape (bool): If True, add absolute position embedding to the patch embedding. Default: False + patch_norm (bool): If True, add normalization after patch embedding. Default: True + use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False + upscale: Upscale factor. 2/3/4/8 for image SR, 1 for denoising and compress artifact reduction + img_range: Image range. 1. or 255. + upsampler: The reconstruction reconstruction module. 'pixelshuffle'/'pixelshuffledirect'/'nearest+conv'/None + resi_connection: The convolutional block before residual connection. '1conv'/'3conv' + """ + + def __init__(self, img_size=64, patch_size=1, in_chans=3, + embed_dim=96, depths=(6, 6, 6, 6), num_heads=(6, 6, 6, 6), + window_size=7, mlp_ratio=4., qkv_bias=True, qk_scale=None, + drop_rate=0., attn_drop_rate=0., drop_path_rate=0.1, + norm_layer=nn.LayerNorm, ape=False, patch_norm=True, + use_checkpoint=False, upscale=2, img_range=1., upsampler='', resi_connection='1conv', + **kwargs): + super(SwinIR, self).__init__() + num_in_ch = in_chans + num_out_ch = in_chans + num_feat = 64 + self.img_range = img_range + if in_chans == 3: + rgb_mean = (0.4488, 0.4371, 0.4040) + self.mean = torch.Tensor(rgb_mean).view(1, 3, 1, 1) + else: + self.mean = torch.zeros(1, 1, 1, 1) + self.upscale = upscale + self.upsampler = upsampler + self.window_size = window_size + + ##################################################################################################### + ################################### 1, shallow feature extraction ################################### + self.conv_first = nn.Conv2d(num_in_ch, embed_dim, 3, 1, 1) + + ##################################################################################################### + ################################### 2, deep feature extraction ###################################### + self.num_layers = len(depths) + self.embed_dim = embed_dim + self.ape = ape + self.patch_norm = patch_norm + self.num_features = embed_dim + self.mlp_ratio = mlp_ratio + + # split image into non-overlapping patches + self.patch_embed = PatchEmbed( + img_size=img_size, patch_size=patch_size, in_chans=embed_dim, embed_dim=embed_dim, + norm_layer=norm_layer if self.patch_norm else None) + num_patches = self.patch_embed.num_patches + patches_resolution = self.patch_embed.patches_resolution + self.patches_resolution = patches_resolution + + # merge non-overlapping patches into image + self.patch_unembed = PatchUnEmbed( + img_size=img_size, patch_size=patch_size, in_chans=embed_dim, embed_dim=embed_dim, + norm_layer=norm_layer if self.patch_norm else None) + + # absolute position embedding + if self.ape: + self.absolute_pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim)) + trunc_normal_(self.absolute_pos_embed, std=.02) + + self.pos_drop = nn.Dropout(p=drop_rate) + + # stochastic depth + dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] # stochastic depth decay rule + + # build Residual Swin Transformer blocks (RSTB) + self.layers = nn.ModuleList() + for i_layer in range(self.num_layers): + layer = RSTB(dim=embed_dim, + input_resolution=(patches_resolution[0], + patches_resolution[1]), + depth=depths[i_layer], + num_heads=num_heads[i_layer], + window_size=window_size, + mlp_ratio=self.mlp_ratio, + qkv_bias=qkv_bias, qk_scale=qk_scale, + drop=drop_rate, attn_drop=attn_drop_rate, + drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])], # no impact on SR results + norm_layer=norm_layer, + downsample=None, + use_checkpoint=use_checkpoint, + img_size=img_size, + patch_size=patch_size, + resi_connection=resi_connection + + ) + self.layers.append(layer) + self.norm = norm_layer(self.num_features) + + # build the last conv layer in deep feature extraction + if resi_connection == '1conv': + self.conv_after_body = nn.Conv2d(embed_dim, embed_dim, 3, 1, 1) + elif resi_connection == '3conv': + # to save parameters and memory + self.conv_after_body = nn.Sequential(nn.Conv2d(embed_dim, embed_dim // 4, 3, 1, 1), + nn.LeakyReLU(negative_slope=0.2, inplace=True), + nn.Conv2d(embed_dim // 4, embed_dim // 4, 1, 1, 0), + nn.LeakyReLU(negative_slope=0.2, inplace=True), + nn.Conv2d(embed_dim // 4, embed_dim, 3, 1, 1)) + + ##################################################################################################### + ################################ 3, high quality image reconstruction ################################ + if self.upsampler == 'pixelshuffle': + # for classical SR + self.conv_before_upsample = nn.Sequential(nn.Conv2d(embed_dim, num_feat, 3, 1, 1), + nn.LeakyReLU(inplace=True)) + self.upsample = Upsample(upscale, num_feat) + self.conv_last = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1) + elif self.upsampler == 'pixelshuffledirect': + # for lightweight SR (to save parameters) + self.upsample = UpsampleOneStep(upscale, embed_dim, num_out_ch, + (patches_resolution[0], patches_resolution[1])) + elif self.upsampler == 'nearest+conv': + # for real-world SR (less artifacts) + self.conv_before_upsample = nn.Sequential(nn.Conv2d(embed_dim, num_feat, 3, 1, 1), + nn.LeakyReLU(inplace=True)) + self.conv_up1 = nn.Conv2d(num_feat, num_feat, 3, 1, 1) + if self.upscale == 4: + self.conv_up2 = nn.Conv2d(num_feat, num_feat, 3, 1, 1) + self.conv_hr = nn.Conv2d(num_feat, num_feat, 3, 1, 1) + self.conv_last = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1) + self.lrelu = nn.LeakyReLU(negative_slope=0.2, inplace=True) + else: + # for image denoising and JPEG compression artifact reduction + self.conv_last = nn.Conv2d(embed_dim, num_out_ch, 3, 1, 1) + + self.apply(self._init_weights) + + def _init_weights(self, m): + if isinstance(m, nn.Linear): + trunc_normal_(m.weight, std=.02) + if isinstance(m, nn.Linear) and m.bias is not None: + nn.init.constant_(m.bias, 0) + elif isinstance(m, nn.LayerNorm): + nn.init.constant_(m.bias, 0) + nn.init.constant_(m.weight, 1.0) + + @torch.jit.ignore + def no_weight_decay(self): + return {'absolute_pos_embed'} + + @torch.jit.ignore + def no_weight_decay_keywords(self): + return {'relative_position_bias_table'} + + def check_image_size(self, x): + _, _, h, w = x.size() + mod_pad_h = (self.window_size - h % self.window_size) % self.window_size + mod_pad_w = (self.window_size - w % self.window_size) % self.window_size + x = F.pad(x, (0, mod_pad_w, 0, mod_pad_h), 'reflect') + return x + + def forward_features(self, x): + x_size = (x.shape[2], x.shape[3]) + x = self.patch_embed(x) + if self.ape: + x = x + self.absolute_pos_embed + x = self.pos_drop(x) + + for layer in self.layers: + x = layer(x, x_size) + + x = self.norm(x) # B L C + x = self.patch_unembed(x, x_size) + + return x + + def forward(self, x): + H, W = x.shape[2:] + x = self.check_image_size(x) + + self.mean = self.mean.type_as(x) + x = (x - self.mean) * self.img_range + + if self.upsampler == 'pixelshuffle': + # for classical SR + x = self.conv_first(x) + x = self.conv_after_body(self.forward_features(x)) + x + x = self.conv_before_upsample(x) + x = self.conv_last(self.upsample(x)) + elif self.upsampler == 'pixelshuffledirect': + # for lightweight SR + x = self.conv_first(x) + x = self.conv_after_body(self.forward_features(x)) + x + x = self.upsample(x) + elif self.upsampler == 'nearest+conv': + # for real-world SR + x = self.conv_first(x) + x = self.conv_after_body(self.forward_features(x)) + x + x = self.conv_before_upsample(x) + x = self.lrelu(self.conv_up1(torch.nn.functional.interpolate(x, scale_factor=2, mode='nearest'))) + if self.upscale == 4: + x = self.lrelu(self.conv_up2(torch.nn.functional.interpolate(x, scale_factor=2, mode='nearest'))) + x = self.conv_last(self.lrelu(self.conv_hr(x))) + else: + # for image denoising and JPEG compression artifact reduction + x_first = self.conv_first(x) + res = self.conv_after_body(self.forward_features(x_first)) + x_first + x = x + self.conv_last(res) + + x = x / self.img_range + self.mean + + return x[:, :, :H*self.upscale, :W*self.upscale] + + def flops(self): + flops = 0 + H, W = self.patches_resolution + flops += H * W * 3 * self.embed_dim * 9 + flops += self.patch_embed.flops() + for layer in self.layers: + flops += layer.flops() + flops += H * W * 3 * self.embed_dim * self.embed_dim + flops += self.upsample.flops() + return flops + + +if __name__ == '__main__': + upscale = 4 + window_size = 8 + height = (1024 // upscale // window_size + 1) * window_size + width = (720 // upscale // window_size + 1) * window_size + model = SwinIR(upscale=2, img_size=(height, width), + window_size=window_size, img_range=1., depths=[6, 6, 6, 6], + embed_dim=60, num_heads=[6, 6, 6, 6], mlp_ratio=2, upsampler='pixelshuffledirect') + print(model) + print(height, width, model.flops() / 1e9) + + x = torch.randn((1, 3, height, width)) + x = model(x) + print(x.shape) diff --git a/extensions-builtin/SwinIR/swinir_model_arch_v2.py b/extensions-builtin/SwinIR/swinir_model_arch_v2.py new file mode 100644 index 0000000000000000000000000000000000000000..59219f69a9a7f8365628cb2f4f57f5cd0104147a --- /dev/null +++ b/extensions-builtin/SwinIR/swinir_model_arch_v2.py @@ -0,0 +1,1017 @@ +# ----------------------------------------------------------------------------------- +# Swin2SR: Swin2SR: SwinV2 Transformer for Compressed Image Super-Resolution and Restoration, https://arxiv.org/abs/ +# Written by Conde and Choi et al. +# ----------------------------------------------------------------------------------- + +import math +import numpy as np +import torch +import torch.nn as nn +import torch.nn.functional as F +import torch.utils.checkpoint as checkpoint +from timm.models.layers import DropPath, to_2tuple, trunc_normal_ + + +class Mlp(nn.Module): + def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.): + super().__init__() + out_features = out_features or in_features + hidden_features = hidden_features or in_features + self.fc1 = nn.Linear(in_features, hidden_features) + self.act = act_layer() + self.fc2 = nn.Linear(hidden_features, out_features) + self.drop = nn.Dropout(drop) + + def forward(self, x): + x = self.fc1(x) + x = self.act(x) + x = self.drop(x) + x = self.fc2(x) + x = self.drop(x) + return x + + +def window_partition(x, window_size): + """ + Args: + x: (B, H, W, C) + window_size (int): window size + Returns: + windows: (num_windows*B, window_size, window_size, C) + """ + B, H, W, C = x.shape + x = x.view(B, H // window_size, window_size, W // window_size, window_size, C) + windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C) + return windows + + +def window_reverse(windows, window_size, H, W): + """ + Args: + windows: (num_windows*B, window_size, window_size, C) + window_size (int): Window size + H (int): Height of image + W (int): Width of image + Returns: + x: (B, H, W, C) + """ + B = int(windows.shape[0] / (H * W / window_size / window_size)) + x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1) + x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1) + return x + +class WindowAttention(nn.Module): + r""" Window based multi-head self attention (W-MSA) module with relative position bias. + It supports both of shifted and non-shifted window. + Args: + dim (int): Number of input channels. + window_size (tuple[int]): The height and width of the window. + num_heads (int): Number of attention heads. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0 + proj_drop (float, optional): Dropout ratio of output. Default: 0.0 + pretrained_window_size (tuple[int]): The height and width of the window in pre-training. + """ + + def __init__(self, dim, window_size, num_heads, qkv_bias=True, attn_drop=0., proj_drop=0., + pretrained_window_size=(0, 0)): + + super().__init__() + self.dim = dim + self.window_size = window_size # Wh, Ww + self.pretrained_window_size = pretrained_window_size + self.num_heads = num_heads + + self.logit_scale = nn.Parameter(torch.log(10 * torch.ones((num_heads, 1, 1))), requires_grad=True) + + # mlp to generate continuous relative position bias + self.cpb_mlp = nn.Sequential(nn.Linear(2, 512, bias=True), + nn.ReLU(inplace=True), + nn.Linear(512, num_heads, bias=False)) + + # get relative_coords_table + relative_coords_h = torch.arange(-(self.window_size[0] - 1), self.window_size[0], dtype=torch.float32) + relative_coords_w = torch.arange(-(self.window_size[1] - 1), self.window_size[1], dtype=torch.float32) + relative_coords_table = torch.stack( + torch.meshgrid([relative_coords_h, + relative_coords_w])).permute(1, 2, 0).contiguous().unsqueeze(0) # 1, 2*Wh-1, 2*Ww-1, 2 + if pretrained_window_size[0] > 0: + relative_coords_table[:, :, :, 0] /= (pretrained_window_size[0] - 1) + relative_coords_table[:, :, :, 1] /= (pretrained_window_size[1] - 1) + else: + relative_coords_table[:, :, :, 0] /= (self.window_size[0] - 1) + relative_coords_table[:, :, :, 1] /= (self.window_size[1] - 1) + relative_coords_table *= 8 # normalize to -8, 8 + relative_coords_table = torch.sign(relative_coords_table) * torch.log2( + torch.abs(relative_coords_table) + 1.0) / np.log2(8) + + self.register_buffer("relative_coords_table", relative_coords_table) + + # get pair-wise relative position index for each token inside the window + coords_h = torch.arange(self.window_size[0]) + coords_w = torch.arange(self.window_size[1]) + coords = torch.stack(torch.meshgrid([coords_h, coords_w])) # 2, Wh, Ww + coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww + relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww + relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2 + relative_coords[:, :, 0] += self.window_size[0] - 1 # shift to start from 0 + relative_coords[:, :, 1] += self.window_size[1] - 1 + relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1 + relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww + self.register_buffer("relative_position_index", relative_position_index) + + self.qkv = nn.Linear(dim, dim * 3, bias=False) + if qkv_bias: + self.q_bias = nn.Parameter(torch.zeros(dim)) + self.v_bias = nn.Parameter(torch.zeros(dim)) + else: + self.q_bias = None + self.v_bias = None + self.attn_drop = nn.Dropout(attn_drop) + self.proj = nn.Linear(dim, dim) + self.proj_drop = nn.Dropout(proj_drop) + self.softmax = nn.Softmax(dim=-1) + + def forward(self, x, mask=None): + """ + Args: + x: input features with shape of (num_windows*B, N, C) + mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None + """ + B_, N, C = x.shape + qkv_bias = None + if self.q_bias is not None: + qkv_bias = torch.cat((self.q_bias, torch.zeros_like(self.v_bias, requires_grad=False), self.v_bias)) + qkv = F.linear(input=x, weight=self.qkv.weight, bias=qkv_bias) + qkv = qkv.reshape(B_, N, 3, self.num_heads, -1).permute(2, 0, 3, 1, 4) + q, k, v = qkv[0], qkv[1], qkv[2] # make torchscript happy (cannot use tensor as tuple) + + # cosine attention + attn = (F.normalize(q, dim=-1) @ F.normalize(k, dim=-1).transpose(-2, -1)) + logit_scale = torch.clamp(self.logit_scale, max=torch.log(torch.tensor(1. / 0.01)).to(self.logit_scale.device)).exp() + attn = attn * logit_scale + + relative_position_bias_table = self.cpb_mlp(self.relative_coords_table).view(-1, self.num_heads) + relative_position_bias = relative_position_bias_table[self.relative_position_index.view(-1)].view( + self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1) # Wh*Ww,Wh*Ww,nH + relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous() # nH, Wh*Ww, Wh*Ww + relative_position_bias = 16 * torch.sigmoid(relative_position_bias) + attn = attn + relative_position_bias.unsqueeze(0) + + if mask is not None: + nW = mask.shape[0] + attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0) + attn = attn.view(-1, self.num_heads, N, N) + attn = self.softmax(attn) + else: + attn = self.softmax(attn) + + attn = self.attn_drop(attn) + + x = (attn @ v).transpose(1, 2).reshape(B_, N, C) + x = self.proj(x) + x = self.proj_drop(x) + return x + + def extra_repr(self) -> str: + return f'dim={self.dim}, window_size={self.window_size}, ' \ + f'pretrained_window_size={self.pretrained_window_size}, num_heads={self.num_heads}' + + def flops(self, N): + # calculate flops for 1 window with token length of N + flops = 0 + # qkv = self.qkv(x) + flops += N * self.dim * 3 * self.dim + # attn = (q @ k.transpose(-2, -1)) + flops += self.num_heads * N * (self.dim // self.num_heads) * N + # x = (attn @ v) + flops += self.num_heads * N * N * (self.dim // self.num_heads) + # x = self.proj(x) + flops += N * self.dim * self.dim + return flops + +class SwinTransformerBlock(nn.Module): + r""" Swin Transformer Block. + Args: + dim (int): Number of input channels. + input_resolution (tuple[int]): Input resulotion. + num_heads (int): Number of attention heads. + window_size (int): Window size. + shift_size (int): Shift size for SW-MSA. + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + drop (float, optional): Dropout rate. Default: 0.0 + attn_drop (float, optional): Attention dropout rate. Default: 0.0 + drop_path (float, optional): Stochastic depth rate. Default: 0.0 + act_layer (nn.Module, optional): Activation layer. Default: nn.GELU + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + pretrained_window_size (int): Window size in pre-training. + """ + + def __init__(self, dim, input_resolution, num_heads, window_size=7, shift_size=0, + mlp_ratio=4., qkv_bias=True, drop=0., attn_drop=0., drop_path=0., + act_layer=nn.GELU, norm_layer=nn.LayerNorm, pretrained_window_size=0): + super().__init__() + self.dim = dim + self.input_resolution = input_resolution + self.num_heads = num_heads + self.window_size = window_size + self.shift_size = shift_size + self.mlp_ratio = mlp_ratio + if min(self.input_resolution) <= self.window_size: + # if window size is larger than input resolution, we don't partition windows + self.shift_size = 0 + self.window_size = min(self.input_resolution) + assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size" + + self.norm1 = norm_layer(dim) + self.attn = WindowAttention( + dim, window_size=to_2tuple(self.window_size), num_heads=num_heads, + qkv_bias=qkv_bias, attn_drop=attn_drop, proj_drop=drop, + pretrained_window_size=to_2tuple(pretrained_window_size)) + + self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity() + self.norm2 = norm_layer(dim) + mlp_hidden_dim = int(dim * mlp_ratio) + self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop) + + if self.shift_size > 0: + attn_mask = self.calculate_mask(self.input_resolution) + else: + attn_mask = None + + self.register_buffer("attn_mask", attn_mask) + + def calculate_mask(self, x_size): + # calculate attention mask for SW-MSA + H, W = x_size + img_mask = torch.zeros((1, H, W, 1)) # 1 H W 1 + h_slices = (slice(0, -self.window_size), + slice(-self.window_size, -self.shift_size), + slice(-self.shift_size, None)) + w_slices = (slice(0, -self.window_size), + slice(-self.window_size, -self.shift_size), + slice(-self.shift_size, None)) + cnt = 0 + for h in h_slices: + for w in w_slices: + img_mask[:, h, w, :] = cnt + cnt += 1 + + mask_windows = window_partition(img_mask, self.window_size) # nW, window_size, window_size, 1 + mask_windows = mask_windows.view(-1, self.window_size * self.window_size) + attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2) + attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0)) + + return attn_mask + + def forward(self, x, x_size): + H, W = x_size + B, L, C = x.shape + #assert L == H * W, "input feature has wrong size" + + shortcut = x + x = x.view(B, H, W, C) + + # cyclic shift + if self.shift_size > 0: + shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2)) + else: + shifted_x = x + + # partition windows + x_windows = window_partition(shifted_x, self.window_size) # nW*B, window_size, window_size, C + x_windows = x_windows.view(-1, self.window_size * self.window_size, C) # nW*B, window_size*window_size, C + + # W-MSA/SW-MSA (to be compatible for testing on images whose shapes are the multiple of window size + if self.input_resolution == x_size: + attn_windows = self.attn(x_windows, mask=self.attn_mask) # nW*B, window_size*window_size, C + else: + attn_windows = self.attn(x_windows, mask=self.calculate_mask(x_size).to(x.device)) + + # merge windows + attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C) + shifted_x = window_reverse(attn_windows, self.window_size, H, W) # B H' W' C + + # reverse cyclic shift + if self.shift_size > 0: + x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2)) + else: + x = shifted_x + x = x.view(B, H * W, C) + x = shortcut + self.drop_path(self.norm1(x)) + + # FFN + x = x + self.drop_path(self.norm2(self.mlp(x))) + + return x + + def extra_repr(self) -> str: + return f"dim={self.dim}, input_resolution={self.input_resolution}, num_heads={self.num_heads}, " \ + f"window_size={self.window_size}, shift_size={self.shift_size}, mlp_ratio={self.mlp_ratio}" + + def flops(self): + flops = 0 + H, W = self.input_resolution + # norm1 + flops += self.dim * H * W + # W-MSA/SW-MSA + nW = H * W / self.window_size / self.window_size + flops += nW * self.attn.flops(self.window_size * self.window_size) + # mlp + flops += 2 * H * W * self.dim * self.dim * self.mlp_ratio + # norm2 + flops += self.dim * H * W + return flops + +class PatchMerging(nn.Module): + r""" Patch Merging Layer. + Args: + input_resolution (tuple[int]): Resolution of input feature. + dim (int): Number of input channels. + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + """ + + def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm): + super().__init__() + self.input_resolution = input_resolution + self.dim = dim + self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False) + self.norm = norm_layer(2 * dim) + + def forward(self, x): + """ + x: B, H*W, C + """ + H, W = self.input_resolution + B, L, C = x.shape + assert L == H * W, "input feature has wrong size" + assert H % 2 == 0 and W % 2 == 0, f"x size ({H}*{W}) are not even." + + x = x.view(B, H, W, C) + + x0 = x[:, 0::2, 0::2, :] # B H/2 W/2 C + x1 = x[:, 1::2, 0::2, :] # B H/2 W/2 C + x2 = x[:, 0::2, 1::2, :] # B H/2 W/2 C + x3 = x[:, 1::2, 1::2, :] # B H/2 W/2 C + x = torch.cat([x0, x1, x2, x3], -1) # B H/2 W/2 4*C + x = x.view(B, -1, 4 * C) # B H/2*W/2 4*C + + x = self.reduction(x) + x = self.norm(x) + + return x + + def extra_repr(self) -> str: + return f"input_resolution={self.input_resolution}, dim={self.dim}" + + def flops(self): + H, W = self.input_resolution + flops = (H // 2) * (W // 2) * 4 * self.dim * 2 * self.dim + flops += H * W * self.dim // 2 + return flops + +class BasicLayer(nn.Module): + """ A basic Swin Transformer layer for one stage. + Args: + dim (int): Number of input channels. + input_resolution (tuple[int]): Input resolution. + depth (int): Number of blocks. + num_heads (int): Number of attention heads. + window_size (int): Local window size. + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + drop (float, optional): Dropout rate. Default: 0.0 + attn_drop (float, optional): Attention dropout rate. Default: 0.0 + drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0 + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None + use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False. + pretrained_window_size (int): Local window size in pre-training. + """ + + def __init__(self, dim, input_resolution, depth, num_heads, window_size, + mlp_ratio=4., qkv_bias=True, drop=0., attn_drop=0., + drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False, + pretrained_window_size=0): + + super().__init__() + self.dim = dim + self.input_resolution = input_resolution + self.depth = depth + self.use_checkpoint = use_checkpoint + + # build blocks + self.blocks = nn.ModuleList([ + SwinTransformerBlock(dim=dim, input_resolution=input_resolution, + num_heads=num_heads, window_size=window_size, + shift_size=0 if (i % 2 == 0) else window_size // 2, + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, + drop=drop, attn_drop=attn_drop, + drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path, + norm_layer=norm_layer, + pretrained_window_size=pretrained_window_size) + for i in range(depth)]) + + # patch merging layer + if downsample is not None: + self.downsample = downsample(input_resolution, dim=dim, norm_layer=norm_layer) + else: + self.downsample = None + + def forward(self, x, x_size): + for blk in self.blocks: + if self.use_checkpoint: + x = checkpoint.checkpoint(blk, x, x_size) + else: + x = blk(x, x_size) + if self.downsample is not None: + x = self.downsample(x) + return x + + def extra_repr(self) -> str: + return f"dim={self.dim}, input_resolution={self.input_resolution}, depth={self.depth}" + + def flops(self): + flops = 0 + for blk in self.blocks: + flops += blk.flops() + if self.downsample is not None: + flops += self.downsample.flops() + return flops + + def _init_respostnorm(self): + for blk in self.blocks: + nn.init.constant_(blk.norm1.bias, 0) + nn.init.constant_(blk.norm1.weight, 0) + nn.init.constant_(blk.norm2.bias, 0) + nn.init.constant_(blk.norm2.weight, 0) + +class PatchEmbed(nn.Module): + r""" Image to Patch Embedding + Args: + img_size (int): Image size. Default: 224. + patch_size (int): Patch token size. Default: 4. + in_chans (int): Number of input image channels. Default: 3. + embed_dim (int): Number of linear projection output channels. Default: 96. + norm_layer (nn.Module, optional): Normalization layer. Default: None + """ + + def __init__(self, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None): + super().__init__() + img_size = to_2tuple(img_size) + patch_size = to_2tuple(patch_size) + patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]] + self.img_size = img_size + self.patch_size = patch_size + self.patches_resolution = patches_resolution + self.num_patches = patches_resolution[0] * patches_resolution[1] + + self.in_chans = in_chans + self.embed_dim = embed_dim + + self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size) + if norm_layer is not None: + self.norm = norm_layer(embed_dim) + else: + self.norm = None + + def forward(self, x): + B, C, H, W = x.shape + # FIXME look at relaxing size constraints + # assert H == self.img_size[0] and W == self.img_size[1], + # f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})." + x = self.proj(x).flatten(2).transpose(1, 2) # B Ph*Pw C + if self.norm is not None: + x = self.norm(x) + return x + + def flops(self): + Ho, Wo = self.patches_resolution + flops = Ho * Wo * self.embed_dim * self.in_chans * (self.patch_size[0] * self.patch_size[1]) + if self.norm is not None: + flops += Ho * Wo * self.embed_dim + return flops + +class RSTB(nn.Module): + """Residual Swin Transformer Block (RSTB). + + Args: + dim (int): Number of input channels. + input_resolution (tuple[int]): Input resolution. + depth (int): Number of blocks. + num_heads (int): Number of attention heads. + window_size (int): Local window size. + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. + qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True + drop (float, optional): Dropout rate. Default: 0.0 + attn_drop (float, optional): Attention dropout rate. Default: 0.0 + drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0 + norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm + downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None + use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False. + img_size: Input image size. + patch_size: Patch size. + resi_connection: The convolutional block before residual connection. + """ + + def __init__(self, dim, input_resolution, depth, num_heads, window_size, + mlp_ratio=4., qkv_bias=True, drop=0., attn_drop=0., + drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False, + img_size=224, patch_size=4, resi_connection='1conv'): + super(RSTB, self).__init__() + + self.dim = dim + self.input_resolution = input_resolution + + self.residual_group = BasicLayer(dim=dim, + input_resolution=input_resolution, + depth=depth, + num_heads=num_heads, + window_size=window_size, + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, + drop=drop, attn_drop=attn_drop, + drop_path=drop_path, + norm_layer=norm_layer, + downsample=downsample, + use_checkpoint=use_checkpoint) + + if resi_connection == '1conv': + self.conv = nn.Conv2d(dim, dim, 3, 1, 1) + elif resi_connection == '3conv': + # to save parameters and memory + self.conv = nn.Sequential(nn.Conv2d(dim, dim // 4, 3, 1, 1), nn.LeakyReLU(negative_slope=0.2, inplace=True), + nn.Conv2d(dim // 4, dim // 4, 1, 1, 0), + nn.LeakyReLU(negative_slope=0.2, inplace=True), + nn.Conv2d(dim // 4, dim, 3, 1, 1)) + + self.patch_embed = PatchEmbed( + img_size=img_size, patch_size=patch_size, in_chans=dim, embed_dim=dim, + norm_layer=None) + + self.patch_unembed = PatchUnEmbed( + img_size=img_size, patch_size=patch_size, in_chans=dim, embed_dim=dim, + norm_layer=None) + + def forward(self, x, x_size): + return self.patch_embed(self.conv(self.patch_unembed(self.residual_group(x, x_size), x_size))) + x + + def flops(self): + flops = 0 + flops += self.residual_group.flops() + H, W = self.input_resolution + flops += H * W * self.dim * self.dim * 9 + flops += self.patch_embed.flops() + flops += self.patch_unembed.flops() + + return flops + +class PatchUnEmbed(nn.Module): + r""" Image to Patch Unembedding + + Args: + img_size (int): Image size. Default: 224. + patch_size (int): Patch token size. Default: 4. + in_chans (int): Number of input image channels. Default: 3. + embed_dim (int): Number of linear projection output channels. Default: 96. + norm_layer (nn.Module, optional): Normalization layer. Default: None + """ + + def __init__(self, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None): + super().__init__() + img_size = to_2tuple(img_size) + patch_size = to_2tuple(patch_size) + patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]] + self.img_size = img_size + self.patch_size = patch_size + self.patches_resolution = patches_resolution + self.num_patches = patches_resolution[0] * patches_resolution[1] + + self.in_chans = in_chans + self.embed_dim = embed_dim + + def forward(self, x, x_size): + B, HW, C = x.shape + x = x.transpose(1, 2).view(B, self.embed_dim, x_size[0], x_size[1]) # B Ph*Pw C + return x + + def flops(self): + flops = 0 + return flops + + +class Upsample(nn.Sequential): + """Upsample module. + + Args: + scale (int): Scale factor. Supported scales: 2^n and 3. + num_feat (int): Channel number of intermediate features. + """ + + def __init__(self, scale, num_feat): + m = [] + if (scale & (scale - 1)) == 0: # scale = 2^n + for _ in range(int(math.log(scale, 2))): + m.append(nn.Conv2d(num_feat, 4 * num_feat, 3, 1, 1)) + m.append(nn.PixelShuffle(2)) + elif scale == 3: + m.append(nn.Conv2d(num_feat, 9 * num_feat, 3, 1, 1)) + m.append(nn.PixelShuffle(3)) + else: + raise ValueError(f'scale {scale} is not supported. ' 'Supported scales: 2^n and 3.') + super(Upsample, self).__init__(*m) + +class Upsample_hf(nn.Sequential): + """Upsample module. + + Args: + scale (int): Scale factor. Supported scales: 2^n and 3. + num_feat (int): Channel number of intermediate features. + """ + + def __init__(self, scale, num_feat): + m = [] + if (scale & (scale - 1)) == 0: # scale = 2^n + for _ in range(int(math.log(scale, 2))): + m.append(nn.Conv2d(num_feat, 4 * num_feat, 3, 1, 1)) + m.append(nn.PixelShuffle(2)) + elif scale == 3: + m.append(nn.Conv2d(num_feat, 9 * num_feat, 3, 1, 1)) + m.append(nn.PixelShuffle(3)) + else: + raise ValueError(f'scale {scale} is not supported. ' 'Supported scales: 2^n and 3.') + super(Upsample_hf, self).__init__(*m) + + +class UpsampleOneStep(nn.Sequential): + """UpsampleOneStep module (the difference with Upsample is that it always only has 1conv + 1pixelshuffle) + Used in lightweight SR to save parameters. + + Args: + scale (int): Scale factor. Supported scales: 2^n and 3. + num_feat (int): Channel number of intermediate features. + + """ + + def __init__(self, scale, num_feat, num_out_ch, input_resolution=None): + self.num_feat = num_feat + self.input_resolution = input_resolution + m = [] + m.append(nn.Conv2d(num_feat, (scale ** 2) * num_out_ch, 3, 1, 1)) + m.append(nn.PixelShuffle(scale)) + super(UpsampleOneStep, self).__init__(*m) + + def flops(self): + H, W = self.input_resolution + flops = H * W * self.num_feat * 3 * 9 + return flops + + + +class Swin2SR(nn.Module): + r""" Swin2SR + A PyTorch impl of : `Swin2SR: SwinV2 Transformer for Compressed Image Super-Resolution and Restoration`. + + Args: + img_size (int | tuple(int)): Input image size. Default 64 + patch_size (int | tuple(int)): Patch size. Default: 1 + in_chans (int): Number of input image channels. Default: 3 + embed_dim (int): Patch embedding dimension. Default: 96 + depths (tuple(int)): Depth of each Swin Transformer layer. + num_heads (tuple(int)): Number of attention heads in different layers. + window_size (int): Window size. Default: 7 + mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4 + qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True + drop_rate (float): Dropout rate. Default: 0 + attn_drop_rate (float): Attention dropout rate. Default: 0 + drop_path_rate (float): Stochastic depth rate. Default: 0.1 + norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm. + ape (bool): If True, add absolute position embedding to the patch embedding. Default: False + patch_norm (bool): If True, add normalization after patch embedding. Default: True + use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False + upscale: Upscale factor. 2/3/4/8 for image SR, 1 for denoising and compress artifact reduction + img_range: Image range. 1. or 255. + upsampler: The reconstruction reconstruction module. 'pixelshuffle'/'pixelshuffledirect'/'nearest+conv'/None + resi_connection: The convolutional block before residual connection. '1conv'/'3conv' + """ + + def __init__(self, img_size=64, patch_size=1, in_chans=3, + embed_dim=96, depths=(6, 6, 6, 6), num_heads=(6, 6, 6, 6), + window_size=7, mlp_ratio=4., qkv_bias=True, + drop_rate=0., attn_drop_rate=0., drop_path_rate=0.1, + norm_layer=nn.LayerNorm, ape=False, patch_norm=True, + use_checkpoint=False, upscale=2, img_range=1., upsampler='', resi_connection='1conv', + **kwargs): + super(Swin2SR, self).__init__() + num_in_ch = in_chans + num_out_ch = in_chans + num_feat = 64 + self.img_range = img_range + if in_chans == 3: + rgb_mean = (0.4488, 0.4371, 0.4040) + self.mean = torch.Tensor(rgb_mean).view(1, 3, 1, 1) + else: + self.mean = torch.zeros(1, 1, 1, 1) + self.upscale = upscale + self.upsampler = upsampler + self.window_size = window_size + + ##################################################################################################### + ################################### 1, shallow feature extraction ################################### + self.conv_first = nn.Conv2d(num_in_ch, embed_dim, 3, 1, 1) + + ##################################################################################################### + ################################### 2, deep feature extraction ###################################### + self.num_layers = len(depths) + self.embed_dim = embed_dim + self.ape = ape + self.patch_norm = patch_norm + self.num_features = embed_dim + self.mlp_ratio = mlp_ratio + + # split image into non-overlapping patches + self.patch_embed = PatchEmbed( + img_size=img_size, patch_size=patch_size, in_chans=embed_dim, embed_dim=embed_dim, + norm_layer=norm_layer if self.patch_norm else None) + num_patches = self.patch_embed.num_patches + patches_resolution = self.patch_embed.patches_resolution + self.patches_resolution = patches_resolution + + # merge non-overlapping patches into image + self.patch_unembed = PatchUnEmbed( + img_size=img_size, patch_size=patch_size, in_chans=embed_dim, embed_dim=embed_dim, + norm_layer=norm_layer if self.patch_norm else None) + + # absolute position embedding + if self.ape: + self.absolute_pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim)) + trunc_normal_(self.absolute_pos_embed, std=.02) + + self.pos_drop = nn.Dropout(p=drop_rate) + + # stochastic depth + dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] # stochastic depth decay rule + + # build Residual Swin Transformer blocks (RSTB) + self.layers = nn.ModuleList() + for i_layer in range(self.num_layers): + layer = RSTB(dim=embed_dim, + input_resolution=(patches_resolution[0], + patches_resolution[1]), + depth=depths[i_layer], + num_heads=num_heads[i_layer], + window_size=window_size, + mlp_ratio=self.mlp_ratio, + qkv_bias=qkv_bias, + drop=drop_rate, attn_drop=attn_drop_rate, + drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])], # no impact on SR results + norm_layer=norm_layer, + downsample=None, + use_checkpoint=use_checkpoint, + img_size=img_size, + patch_size=patch_size, + resi_connection=resi_connection + + ) + self.layers.append(layer) + + if self.upsampler == 'pixelshuffle_hf': + self.layers_hf = nn.ModuleList() + for i_layer in range(self.num_layers): + layer = RSTB(dim=embed_dim, + input_resolution=(patches_resolution[0], + patches_resolution[1]), + depth=depths[i_layer], + num_heads=num_heads[i_layer], + window_size=window_size, + mlp_ratio=self.mlp_ratio, + qkv_bias=qkv_bias, + drop=drop_rate, attn_drop=attn_drop_rate, + drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])], # no impact on SR results + norm_layer=norm_layer, + downsample=None, + use_checkpoint=use_checkpoint, + img_size=img_size, + patch_size=patch_size, + resi_connection=resi_connection + + ) + self.layers_hf.append(layer) + + self.norm = norm_layer(self.num_features) + + # build the last conv layer in deep feature extraction + if resi_connection == '1conv': + self.conv_after_body = nn.Conv2d(embed_dim, embed_dim, 3, 1, 1) + elif resi_connection == '3conv': + # to save parameters and memory + self.conv_after_body = nn.Sequential(nn.Conv2d(embed_dim, embed_dim // 4, 3, 1, 1), + nn.LeakyReLU(negative_slope=0.2, inplace=True), + nn.Conv2d(embed_dim // 4, embed_dim // 4, 1, 1, 0), + nn.LeakyReLU(negative_slope=0.2, inplace=True), + nn.Conv2d(embed_dim // 4, embed_dim, 3, 1, 1)) + + ##################################################################################################### + ################################ 3, high quality image reconstruction ################################ + if self.upsampler == 'pixelshuffle': + # for classical SR + self.conv_before_upsample = nn.Sequential(nn.Conv2d(embed_dim, num_feat, 3, 1, 1), + nn.LeakyReLU(inplace=True)) + self.upsample = Upsample(upscale, num_feat) + self.conv_last = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1) + elif self.upsampler == 'pixelshuffle_aux': + self.conv_bicubic = nn.Conv2d(num_in_ch, num_feat, 3, 1, 1) + self.conv_before_upsample = nn.Sequential( + nn.Conv2d(embed_dim, num_feat, 3, 1, 1), + nn.LeakyReLU(inplace=True)) + self.conv_aux = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1) + self.conv_after_aux = nn.Sequential( + nn.Conv2d(3, num_feat, 3, 1, 1), + nn.LeakyReLU(inplace=True)) + self.upsample = Upsample(upscale, num_feat) + self.conv_last = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1) + + elif self.upsampler == 'pixelshuffle_hf': + self.conv_before_upsample = nn.Sequential(nn.Conv2d(embed_dim, num_feat, 3, 1, 1), + nn.LeakyReLU(inplace=True)) + self.upsample = Upsample(upscale, num_feat) + self.upsample_hf = Upsample_hf(upscale, num_feat) + self.conv_last = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1) + self.conv_first_hf = nn.Sequential(nn.Conv2d(num_feat, embed_dim, 3, 1, 1), + nn.LeakyReLU(inplace=True)) + self.conv_after_body_hf = nn.Conv2d(embed_dim, embed_dim, 3, 1, 1) + self.conv_before_upsample_hf = nn.Sequential( + nn.Conv2d(embed_dim, num_feat, 3, 1, 1), + nn.LeakyReLU(inplace=True)) + self.conv_last_hf = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1) + + elif self.upsampler == 'pixelshuffledirect': + # for lightweight SR (to save parameters) + self.upsample = UpsampleOneStep(upscale, embed_dim, num_out_ch, + (patches_resolution[0], patches_resolution[1])) + elif self.upsampler == 'nearest+conv': + # for real-world SR (less artifacts) + assert self.upscale == 4, 'only support x4 now.' + self.conv_before_upsample = nn.Sequential(nn.Conv2d(embed_dim, num_feat, 3, 1, 1), + nn.LeakyReLU(inplace=True)) + self.conv_up1 = nn.Conv2d(num_feat, num_feat, 3, 1, 1) + self.conv_up2 = nn.Conv2d(num_feat, num_feat, 3, 1, 1) + self.conv_hr = nn.Conv2d(num_feat, num_feat, 3, 1, 1) + self.conv_last = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1) + self.lrelu = nn.LeakyReLU(negative_slope=0.2, inplace=True) + else: + # for image denoising and JPEG compression artifact reduction + self.conv_last = nn.Conv2d(embed_dim, num_out_ch, 3, 1, 1) + + self.apply(self._init_weights) + + def _init_weights(self, m): + if isinstance(m, nn.Linear): + trunc_normal_(m.weight, std=.02) + if isinstance(m, nn.Linear) and m.bias is not None: + nn.init.constant_(m.bias, 0) + elif isinstance(m, nn.LayerNorm): + nn.init.constant_(m.bias, 0) + nn.init.constant_(m.weight, 1.0) + + @torch.jit.ignore + def no_weight_decay(self): + return {'absolute_pos_embed'} + + @torch.jit.ignore + def no_weight_decay_keywords(self): + return {'relative_position_bias_table'} + + def check_image_size(self, x): + _, _, h, w = x.size() + mod_pad_h = (self.window_size - h % self.window_size) % self.window_size + mod_pad_w = (self.window_size - w % self.window_size) % self.window_size + x = F.pad(x, (0, mod_pad_w, 0, mod_pad_h), 'reflect') + return x + + def forward_features(self, x): + x_size = (x.shape[2], x.shape[3]) + x = self.patch_embed(x) + if self.ape: + x = x + self.absolute_pos_embed + x = self.pos_drop(x) + + for layer in self.layers: + x = layer(x, x_size) + + x = self.norm(x) # B L C + x = self.patch_unembed(x, x_size) + + return x + + def forward_features_hf(self, x): + x_size = (x.shape[2], x.shape[3]) + x = self.patch_embed(x) + if self.ape: + x = x + self.absolute_pos_embed + x = self.pos_drop(x) + + for layer in self.layers_hf: + x = layer(x, x_size) + + x = self.norm(x) # B L C + x = self.patch_unembed(x, x_size) + + return x + + def forward(self, x): + H, W = x.shape[2:] + x = self.check_image_size(x) + + self.mean = self.mean.type_as(x) + x = (x - self.mean) * self.img_range + + if self.upsampler == 'pixelshuffle': + # for classical SR + x = self.conv_first(x) + x = self.conv_after_body(self.forward_features(x)) + x + x = self.conv_before_upsample(x) + x = self.conv_last(self.upsample(x)) + elif self.upsampler == 'pixelshuffle_aux': + bicubic = F.interpolate(x, size=(H * self.upscale, W * self.upscale), mode='bicubic', align_corners=False) + bicubic = self.conv_bicubic(bicubic) + x = self.conv_first(x) + x = self.conv_after_body(self.forward_features(x)) + x + x = self.conv_before_upsample(x) + aux = self.conv_aux(x) # b, 3, LR_H, LR_W + x = self.conv_after_aux(aux) + x = self.upsample(x)[:, :, :H * self.upscale, :W * self.upscale] + bicubic[:, :, :H * self.upscale, :W * self.upscale] + x = self.conv_last(x) + aux = aux / self.img_range + self.mean + elif self.upsampler == 'pixelshuffle_hf': + # for classical SR with HF + x = self.conv_first(x) + x = self.conv_after_body(self.forward_features(x)) + x + x_before = self.conv_before_upsample(x) + x_out = self.conv_last(self.upsample(x_before)) + + x_hf = self.conv_first_hf(x_before) + x_hf = self.conv_after_body_hf(self.forward_features_hf(x_hf)) + x_hf + x_hf = self.conv_before_upsample_hf(x_hf) + x_hf = self.conv_last_hf(self.upsample_hf(x_hf)) + x = x_out + x_hf + x_hf = x_hf / self.img_range + self.mean + + elif self.upsampler == 'pixelshuffledirect': + # for lightweight SR + x = self.conv_first(x) + x = self.conv_after_body(self.forward_features(x)) + x + x = self.upsample(x) + elif self.upsampler == 'nearest+conv': + # for real-world SR + x = self.conv_first(x) + x = self.conv_after_body(self.forward_features(x)) + x + x = self.conv_before_upsample(x) + x = self.lrelu(self.conv_up1(torch.nn.functional.interpolate(x, scale_factor=2, mode='nearest'))) + x = self.lrelu(self.conv_up2(torch.nn.functional.interpolate(x, scale_factor=2, mode='nearest'))) + x = self.conv_last(self.lrelu(self.conv_hr(x))) + else: + # for image denoising and JPEG compression artifact reduction + x_first = self.conv_first(x) + res = self.conv_after_body(self.forward_features(x_first)) + x_first + x = x + self.conv_last(res) + + x = x / self.img_range + self.mean + if self.upsampler == "pixelshuffle_aux": + return x[:, :, :H*self.upscale, :W*self.upscale], aux + + elif self.upsampler == "pixelshuffle_hf": + x_out = x_out / self.img_range + self.mean + return x_out[:, :, :H*self.upscale, :W*self.upscale], x[:, :, :H*self.upscale, :W*self.upscale], x_hf[:, :, :H*self.upscale, :W*self.upscale] + + else: + return x[:, :, :H*self.upscale, :W*self.upscale] + + def flops(self): + flops = 0 + H, W = self.patches_resolution + flops += H * W * 3 * self.embed_dim * 9 + flops += self.patch_embed.flops() + for layer in self.layers: + flops += layer.flops() + flops += H * W * 3 * self.embed_dim * self.embed_dim + flops += self.upsample.flops() + return flops + + +if __name__ == '__main__': + upscale = 4 + window_size = 8 + height = (1024 // upscale // window_size + 1) * window_size + width = (720 // upscale // window_size + 1) * window_size + model = Swin2SR(upscale=2, img_size=(height, width), + window_size=window_size, img_range=1., depths=[6, 6, 6, 6], + embed_dim=60, num_heads=[6, 6, 6, 6], mlp_ratio=2, upsampler='pixelshuffledirect') + print(model) + print(height, width, model.flops() / 1e9) + + x = torch.randn((1, 3, height, width)) + x = model(x) + print(x.shape) diff --git a/extensions-builtin/prompt-bracket-checker/javascript/prompt-bracket-checker.js b/extensions-builtin/prompt-bracket-checker/javascript/prompt-bracket-checker.js new file mode 100644 index 0000000000000000000000000000000000000000..114cf94ccbf69b473757f2fc46443a39723a9269 --- /dev/null +++ b/extensions-builtin/prompt-bracket-checker/javascript/prompt-bracket-checker.js @@ -0,0 +1,42 @@ +// Stable Diffusion WebUI - Bracket checker +// By Hingashi no Florin/Bwin4L & @akx +// Counts open and closed brackets (round, square, curly) in the prompt and negative prompt text boxes in the txt2img and img2img tabs. +// If there's a mismatch, the keyword counter turns red and if you hover on it, a tooltip tells you what's wrong. + +function checkBrackets(textArea, counterElt) { + var counts = {}; + (textArea.value.match(/[(){}[\]]/g) || []).forEach(bracket => { + counts[bracket] = (counts[bracket] || 0) + 1; + }); + var errors = []; + + function checkPair(open, close, kind) { + if (counts[open] !== counts[close]) { + errors.push( + `${open}...${close} - Detected ${counts[open] || 0} opening and ${counts[close] || 0} closing ${kind}.` + ); + } + } + + checkPair('(', ')', 'round brackets'); + checkPair('[', ']', 'square brackets'); + checkPair('{', '}', 'curly brackets'); + counterElt.title = errors.join('\n'); + counterElt.classList.toggle('error', errors.length !== 0); +} + +function setupBracketChecking(id_prompt, id_counter) { + var textarea = gradioApp().querySelector("#" + id_prompt + " > label > textarea"); + var counter = gradioApp().getElementById(id_counter); + + if (textarea && counter) { + textarea.addEventListener("input", () => checkBrackets(textarea, counter)); + } +} + +onUiLoaded(function() { + setupBracketChecking('txt2img_prompt', 'txt2img_token_counter'); + setupBracketChecking('txt2img_neg_prompt', 'txt2img_negative_token_counter'); + setupBracketChecking('img2img_prompt', 'img2img_token_counter'); + setupBracketChecking('img2img_neg_prompt', 'img2img_negative_token_counter'); +}); diff --git a/extensions/put extensions here.txt b/extensions/put extensions here.txt new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/html/card-no-preview.png b/html/card-no-preview.png new file mode 100644 index 0000000000000000000000000000000000000000..e2beb2692067db56ac5f7bd5bfc3d895d9063371 Binary files /dev/null and b/html/card-no-preview.png differ diff --git a/html/extra-networks-card.html b/html/extra-networks-card.html new file mode 100644 index 0000000000000000000000000000000000000000..2b32e71269c0f4691aa1c9a31e20d8a3bcd720d8 --- /dev/null +++ b/html/extra-networks-card.html @@ -0,0 +1,14 @@ +
+ {background_image} + {metadata_button} +
+
+ + +
+ {name} + {description} +
+
diff --git a/html/extra-networks-no-cards.html b/html/extra-networks-no-cards.html new file mode 100644 index 0000000000000000000000000000000000000000..389358d6c4b383fdc3c5686e029e7b3b1ae9a493 --- /dev/null +++ b/html/extra-networks-no-cards.html @@ -0,0 +1,8 @@ +
+

Nothing here. Add some content to the following directories:

+ +
    +{dirs} +
+
+ diff --git a/html/footer.html b/html/footer.html new file mode 100644 index 0000000000000000000000000000000000000000..f26e32e9304aedb5a55b0b46a913396f16375f7a --- /dev/null +++ b/html/footer.html @@ -0,0 +1,13 @@ +
+ API +  •  + Github +  •  + Gradio +  •  + Reload UI +
+
+
+{versions} +
diff --git a/html/image-update.svg b/html/image-update.svg new file mode 100644 index 0000000000000000000000000000000000000000..3abf12df0f7774c13203e3c49ec3544649df42f4 --- /dev/null +++ b/html/image-update.svg @@ -0,0 +1,7 @@ + + + + + + + diff --git a/html/licenses.html b/html/licenses.html new file mode 100644 index 0000000000000000000000000000000000000000..ca44deddd3663514962493c06a42a38d608c1229 --- /dev/null +++ b/html/licenses.html @@ -0,0 +1,690 @@ + + +

CodeFormer

+Parts of CodeFormer code had to be copied to be compatible with GFPGAN. +
+S-Lab License 1.0
+
+Copyright 2022 S-Lab
+
+Redistribution and use for non-commercial purpose in source and
+binary forms, with or without modification, are permitted provided
+that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in
+   the documentation and/or other materials provided with the
+   distribution.
+
+3. Neither the name of the copyright holder nor the names of its
+   contributors may be used to endorse or promote products derived
+   from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+In the event that redistribution and/or use for commercial purpose in
+source or binary forms, with or without modification is required,
+please contact the contributor(s) of the work.
+
+ + +

ESRGAN

+Code for architecture and reading models copied. +
+MIT License
+
+Copyright (c) 2021 victorca25
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+ +

Real-ESRGAN

+Some code is copied to support ESRGAN models. +
+BSD 3-Clause License
+
+Copyright (c) 2021, Xintao Wang
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+3. Neither the name of the copyright holder nor the names of its
+   contributors may be used to endorse or promote products derived from
+   this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ +

InvokeAI

+Some code for compatibility with OSX is taken from lstein's repository. +
+MIT License
+
+Copyright (c) 2022 InvokeAI Team
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+ +

LDSR

+Code added by contirubtors, most likely copied from this repository. +
+MIT License
+
+Copyright (c) 2022 Machine Vision and Learning Group, LMU Munich
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+ +

CLIP Interrogator

+Some small amounts of code borrowed and reworked. +
+MIT License
+
+Copyright (c) 2022 pharmapsychotic
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+ +

SwinIR

+Code added by contributors, most likely copied from this repository. + +
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [2021] [SwinIR Authors]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+
+ +

Memory Efficient Attention

+The sub-quadratic cross attention optimization uses modified code from the Memory Efficient Attention package that Alex Birch optimized for 3D tensors. This license is updated to reflect that. +
+MIT License
+
+Copyright (c) 2023 Alex Birch
+Copyright (c) 2023 Amin Rezaei
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+ +

Scaled Dot Product Attention

+Some small amounts of code borrowed and reworked. +
+   Copyright 2023 The HuggingFace Team. All rights reserved.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+      http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+
+ +

Curated transformers

+The MPS workaround for nn.Linear on macOS 13.2.X is based on the MPS workaround for nn.Linear created by danieldk for Curated transformers +
+The MIT License (MIT)
+
+Copyright (C) 2021 ExplosionAI GmbH
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.
+
+ +

TAESD

+Tiny AutoEncoder for Stable Diffusion option for live previews +
+MIT License
+
+Copyright (c) 2023 Ollin Boer Bohan
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
\ No newline at end of file diff --git a/javascript/aspectRatioOverlay.js b/javascript/aspectRatioOverlay.js new file mode 100644 index 0000000000000000000000000000000000000000..1c08a1a97d6e296ad41cb30fe69240acfeb54dbc --- /dev/null +++ b/javascript/aspectRatioOverlay.js @@ -0,0 +1,113 @@ + +let currentWidth = null; +let currentHeight = null; +let arFrameTimeout = setTimeout(function() {}, 0); + +function dimensionChange(e, is_width, is_height) { + + if (is_width) { + currentWidth = e.target.value * 1.0; + } + if (is_height) { + currentHeight = e.target.value * 1.0; + } + + var inImg2img = gradioApp().querySelector("#tab_img2img").style.display == "block"; + + if (!inImg2img) { + return; + } + + var targetElement = null; + + var tabIndex = get_tab_index('mode_img2img'); + if (tabIndex == 0) { // img2img + targetElement = gradioApp().querySelector('#img2img_image div[data-testid=image] img'); + } else if (tabIndex == 1) { //Sketch + targetElement = gradioApp().querySelector('#img2img_sketch div[data-testid=image] img'); + } else if (tabIndex == 2) { // Inpaint + targetElement = gradioApp().querySelector('#img2maskimg div[data-testid=image] img'); + } else if (tabIndex == 3) { // Inpaint sketch + targetElement = gradioApp().querySelector('#inpaint_sketch div[data-testid=image] img'); + } + + + if (targetElement) { + + var arPreviewRect = gradioApp().querySelector('#imageARPreview'); + if (!arPreviewRect) { + arPreviewRect = document.createElement('div'); + arPreviewRect.id = "imageARPreview"; + gradioApp().appendChild(arPreviewRect); + } + + + + var viewportOffset = targetElement.getBoundingClientRect(); + + var viewportscale = Math.min(targetElement.clientWidth / targetElement.naturalWidth, targetElement.clientHeight / targetElement.naturalHeight); + + var scaledx = targetElement.naturalWidth * viewportscale; + var scaledy = targetElement.naturalHeight * viewportscale; + + var cleintRectTop = (viewportOffset.top + window.scrollY); + var cleintRectLeft = (viewportOffset.left + window.scrollX); + var cleintRectCentreY = cleintRectTop + (targetElement.clientHeight / 2); + var cleintRectCentreX = cleintRectLeft + (targetElement.clientWidth / 2); + + var arscale = Math.min(scaledx / currentWidth, scaledy / currentHeight); + var arscaledx = currentWidth * arscale; + var arscaledy = currentHeight * arscale; + + var arRectTop = cleintRectCentreY - (arscaledy / 2); + var arRectLeft = cleintRectCentreX - (arscaledx / 2); + var arRectWidth = arscaledx; + var arRectHeight = arscaledy; + + arPreviewRect.style.top = arRectTop + 'px'; + arPreviewRect.style.left = arRectLeft + 'px'; + arPreviewRect.style.width = arRectWidth + 'px'; + arPreviewRect.style.height = arRectHeight + 'px'; + + clearTimeout(arFrameTimeout); + arFrameTimeout = setTimeout(function() { + arPreviewRect.style.display = 'none'; + }, 2000); + + arPreviewRect.style.display = 'block'; + + } + +} + + +onUiUpdate(function() { + var arPreviewRect = gradioApp().querySelector('#imageARPreview'); + if (arPreviewRect) { + arPreviewRect.style.display = 'none'; + } + var tabImg2img = gradioApp().querySelector("#tab_img2img"); + if (tabImg2img) { + var inImg2img = tabImg2img.style.display == "block"; + if (inImg2img) { + let inputs = gradioApp().querySelectorAll('input'); + inputs.forEach(function(e) { + var is_width = e.parentElement.id == "img2img_width"; + var is_height = e.parentElement.id == "img2img_height"; + + if ((is_width || is_height) && !e.classList.contains('scrollwatch')) { + e.addEventListener('input', function(e) { + dimensionChange(e, is_width, is_height); + }); + e.classList.add('scrollwatch'); + } + if (is_width) { + currentWidth = e.value * 1.0; + } + if (is_height) { + currentHeight = e.value * 1.0; + } + }); + } + } +}); diff --git a/javascript/contextMenus.js b/javascript/contextMenus.js new file mode 100644 index 0000000000000000000000000000000000000000..f14af1d427cb03b3d59ecc906600a1fa0e7dd766 --- /dev/null +++ b/javascript/contextMenus.js @@ -0,0 +1,172 @@ + +var contextMenuInit = function() { + let eventListenerApplied = false; + let menuSpecs = new Map(); + + const uid = function() { + return Date.now().toString(36) + Math.random().toString(36).substring(2); + }; + + function showContextMenu(event, element, menuEntries) { + let posx = event.clientX + document.body.scrollLeft + document.documentElement.scrollLeft; + let posy = event.clientY + document.body.scrollTop + document.documentElement.scrollTop; + + let oldMenu = gradioApp().querySelector('#context-menu'); + if (oldMenu) { + oldMenu.remove(); + } + + let baseStyle = window.getComputedStyle(uiCurrentTab); + + const contextMenu = document.createElement('nav'); + contextMenu.id = "context-menu"; + contextMenu.style.background = baseStyle.background; + contextMenu.style.color = baseStyle.color; + contextMenu.style.fontFamily = baseStyle.fontFamily; + contextMenu.style.top = posy + 'px'; + contextMenu.style.left = posx + 'px'; + + + + const contextMenuList = document.createElement('ul'); + contextMenuList.className = 'context-menu-items'; + contextMenu.append(contextMenuList); + + menuEntries.forEach(function(entry) { + let contextMenuEntry = document.createElement('a'); + contextMenuEntry.innerHTML = entry['name']; + contextMenuEntry.addEventListener("click", function() { + entry['func'](); + }); + contextMenuList.append(contextMenuEntry); + + }); + + gradioApp().appendChild(contextMenu); + + let menuWidth = contextMenu.offsetWidth + 4; + let menuHeight = contextMenu.offsetHeight + 4; + + let windowWidth = window.innerWidth; + let windowHeight = window.innerHeight; + + if ((windowWidth - posx) < menuWidth) { + contextMenu.style.left = windowWidth - menuWidth + "px"; + } + + if ((windowHeight - posy) < menuHeight) { + contextMenu.style.top = windowHeight - menuHeight + "px"; + } + + } + + function appendContextMenuOption(targetElementSelector, entryName, entryFunction) { + + var currentItems = menuSpecs.get(targetElementSelector); + + if (!currentItems) { + currentItems = []; + menuSpecs.set(targetElementSelector, currentItems); + } + let newItem = { + id: targetElementSelector + '_' + uid(), + name: entryName, + func: entryFunction, + isNew: true + }; + + currentItems.push(newItem); + return newItem['id']; + } + + function removeContextMenuOption(uid) { + menuSpecs.forEach(function(v) { + let index = -1; + v.forEach(function(e, ei) { + if (e['id'] == uid) { + index = ei; + } + }); + if (index >= 0) { + v.splice(index, 1); + } + }); + } + + function addContextMenuEventListener() { + if (eventListenerApplied) { + return; + } + gradioApp().addEventListener("click", function(e) { + if (!e.isTrusted) { + return; + } + + let oldMenu = gradioApp().querySelector('#context-menu'); + if (oldMenu) { + oldMenu.remove(); + } + }); + gradioApp().addEventListener("contextmenu", function(e) { + let oldMenu = gradioApp().querySelector('#context-menu'); + if (oldMenu) { + oldMenu.remove(); + } + menuSpecs.forEach(function(v, k) { + if (e.composedPath()[0].matches(k)) { + showContextMenu(e, e.composedPath()[0], v); + e.preventDefault(); + } + }); + }); + eventListenerApplied = true; + + } + + return [appendContextMenuOption, removeContextMenuOption, addContextMenuEventListener]; +}; + +var initResponse = contextMenuInit(); +var appendContextMenuOption = initResponse[0]; +var removeContextMenuOption = initResponse[1]; +var addContextMenuEventListener = initResponse[2]; + +(function() { + //Start example Context Menu Items + let generateOnRepeat = function(genbuttonid, interruptbuttonid) { + let genbutton = gradioApp().querySelector(genbuttonid); + let interruptbutton = gradioApp().querySelector(interruptbuttonid); + if (!interruptbutton.offsetParent) { + genbutton.click(); + } + clearInterval(window.generateOnRepeatInterval); + window.generateOnRepeatInterval = setInterval(function() { + if (!interruptbutton.offsetParent) { + genbutton.click(); + } + }, + 500); + }; + + appendContextMenuOption('#txt2img_generate', 'Generate forever', function() { + generateOnRepeat('#txt2img_generate', '#txt2img_interrupt'); + }); + appendContextMenuOption('#img2img_generate', 'Generate forever', function() { + generateOnRepeat('#img2img_generate', '#img2img_interrupt'); + }); + + let cancelGenerateForever = function() { + clearInterval(window.generateOnRepeatInterval); + }; + + appendContextMenuOption('#txt2img_interrupt', 'Cancel generate forever', cancelGenerateForever); + appendContextMenuOption('#txt2img_generate', 'Cancel generate forever', cancelGenerateForever); + appendContextMenuOption('#img2img_interrupt', 'Cancel generate forever', cancelGenerateForever); + appendContextMenuOption('#img2img_generate', 'Cancel generate forever', cancelGenerateForever); + +})(); +//End example Context Menu Items + +onUiUpdate(function() { + addContextMenuEventListener(); +}); diff --git a/javascript/dragdrop.js b/javascript/dragdrop.js new file mode 100644 index 0000000000000000000000000000000000000000..77a24a070312fd9dd58b5d410e1fbad881e3d4ee --- /dev/null +++ b/javascript/dragdrop.js @@ -0,0 +1,101 @@ +// allows drag-dropping files into gradio image elements, and also pasting images from clipboard + +function isValidImageList(files) { + return files && files?.length === 1 && ['image/png', 'image/gif', 'image/jpeg'].includes(files[0].type); +} + +function dropReplaceImage(imgWrap, files) { + if (!isValidImageList(files)) { + return; + } + + const tmpFile = files[0]; + + imgWrap.querySelector('.modify-upload button + button, .touch-none + div button + button')?.click(); + const callback = () => { + const fileInput = imgWrap.querySelector('input[type="file"]'); + if (fileInput) { + if (files.length === 0) { + files = new DataTransfer(); + files.items.add(tmpFile); + fileInput.files = files.files; + } else { + fileInput.files = files; + } + fileInput.dispatchEvent(new Event('change')); + } + }; + + if (imgWrap.closest('#pnginfo_image')) { + // special treatment for PNG Info tab, wait for fetch request to finish + const oldFetch = window.fetch; + window.fetch = async(input, options) => { + const response = await oldFetch(input, options); + if ('api/predict/' === input) { + const content = await response.text(); + window.fetch = oldFetch; + window.requestAnimationFrame(() => callback()); + return new Response(content, { + status: response.status, + statusText: response.statusText, + headers: response.headers + }); + } + return response; + }; + } else { + window.requestAnimationFrame(() => callback()); + } +} + +window.document.addEventListener('dragover', e => { + const target = e.composedPath()[0]; + const imgWrap = target.closest('[data-testid="image"]'); + if (!imgWrap && target.placeholder && target.placeholder.indexOf("Prompt") == -1) { + return; + } + e.stopPropagation(); + e.preventDefault(); + e.dataTransfer.dropEffect = 'copy'; +}); + +window.document.addEventListener('drop', e => { + const target = e.composedPath()[0]; + if (target.placeholder.indexOf("Prompt") == -1) { + return; + } + const imgWrap = target.closest('[data-testid="image"]'); + if (!imgWrap) { + return; + } + e.stopPropagation(); + e.preventDefault(); + const files = e.dataTransfer.files; + dropReplaceImage(imgWrap, files); +}); + +window.addEventListener('paste', e => { + const files = e.clipboardData.files; + if (!isValidImageList(files)) { + return; + } + + const visibleImageFields = [...gradioApp().querySelectorAll('[data-testid="image"]')] + .filter(el => uiElementIsVisible(el)) + .sort((a, b) => uiElementInSight(b) - uiElementInSight(a)); + + + if (!visibleImageFields.length) { + return; + } + + const firstFreeImageField = visibleImageFields + .filter(el => el.querySelector('input[type=file]'))?.[0]; + + dropReplaceImage( + firstFreeImageField ? + firstFreeImageField : + visibleImageFields[visibleImageFields.length - 1] + , files + ); +}); diff --git a/javascript/edit-attention.js b/javascript/edit-attention.js new file mode 100644 index 0000000000000000000000000000000000000000..ffa73147ff2e064f50c1f769956a9c8acea4ca8d --- /dev/null +++ b/javascript/edit-attention.js @@ -0,0 +1,120 @@ +function keyupEditAttention(event) { + let target = event.originalTarget || event.composedPath()[0]; + if (!target.matches("*:is([id*='_toprow'] [id*='_prompt'], .prompt) textarea")) return; + if (!(event.metaKey || event.ctrlKey)) return; + + let isPlus = event.key == "ArrowUp"; + let isMinus = event.key == "ArrowDown"; + if (!isPlus && !isMinus) return; + + let selectionStart = target.selectionStart; + let selectionEnd = target.selectionEnd; + let text = target.value; + + function selectCurrentParenthesisBlock(OPEN, CLOSE) { + if (selectionStart !== selectionEnd) return false; + + // Find opening parenthesis around current cursor + const before = text.substring(0, selectionStart); + let beforeParen = before.lastIndexOf(OPEN); + if (beforeParen == -1) return false; + let beforeParenClose = before.lastIndexOf(CLOSE); + while (beforeParenClose !== -1 && beforeParenClose > beforeParen) { + beforeParen = before.lastIndexOf(OPEN, beforeParen - 1); + beforeParenClose = before.lastIndexOf(CLOSE, beforeParenClose - 1); + } + + // Find closing parenthesis around current cursor + const after = text.substring(selectionStart); + let afterParen = after.indexOf(CLOSE); + if (afterParen == -1) return false; + let afterParenOpen = after.indexOf(OPEN); + while (afterParenOpen !== -1 && afterParen > afterParenOpen) { + afterParen = after.indexOf(CLOSE, afterParen + 1); + afterParenOpen = after.indexOf(OPEN, afterParenOpen + 1); + } + if (beforeParen === -1 || afterParen === -1) return false; + + // Set the selection to the text between the parenthesis + const parenContent = text.substring(beforeParen + 1, selectionStart + afterParen); + const lastColon = parenContent.lastIndexOf(":"); + selectionStart = beforeParen + 1; + selectionEnd = selectionStart + lastColon; + target.setSelectionRange(selectionStart, selectionEnd); + return true; + } + + function selectCurrentWord() { + if (selectionStart !== selectionEnd) return false; + const delimiters = opts.keyedit_delimiters + " \r\n\t"; + + // seek backward until to find beggining + while (!delimiters.includes(text[selectionStart - 1]) && selectionStart > 0) { + selectionStart--; + } + + // seek forward to find end + while (!delimiters.includes(text[selectionEnd]) && selectionEnd < text.length) { + selectionEnd++; + } + + target.setSelectionRange(selectionStart, selectionEnd); + return true; + } + + // If the user hasn't selected anything, let's select their current parenthesis block or word + if (!selectCurrentParenthesisBlock('<', '>') && !selectCurrentParenthesisBlock('(', ')')) { + selectCurrentWord(); + } + + event.preventDefault(); + + var closeCharacter = ')'; + var delta = opts.keyedit_precision_attention; + + if (selectionStart > 0 && text[selectionStart - 1] == '<') { + closeCharacter = '>'; + delta = opts.keyedit_precision_extra; + } else if (selectionStart == 0 || text[selectionStart - 1] != "(") { + + // do not include spaces at the end + while (selectionEnd > selectionStart && text[selectionEnd - 1] == ' ') { + selectionEnd -= 1; + } + if (selectionStart == selectionEnd) { + return; + } + + text = text.slice(0, selectionStart) + "(" + text.slice(selectionStart, selectionEnd) + ":1.0)" + text.slice(selectionEnd); + + selectionStart += 1; + selectionEnd += 1; + } + + var end = text.slice(selectionEnd + 1).indexOf(closeCharacter) + 1; + var weight = parseFloat(text.slice(selectionEnd + 1, selectionEnd + 1 + end)); + if (isNaN(weight)) return; + + weight += isPlus ? delta : -delta; + weight = parseFloat(weight.toPrecision(12)); + if (String(weight).length == 1) weight += ".0"; + + if (closeCharacter == ')' && weight == 1) { + text = text.slice(0, selectionStart - 1) + text.slice(selectionStart, selectionEnd) + text.slice(selectionEnd + 5); + selectionStart--; + selectionEnd--; + } else { + text = text.slice(0, selectionEnd + 1) + weight + text.slice(selectionEnd + 1 + end - 1); + } + + target.focus(); + target.value = text; + target.selectionStart = selectionStart; + target.selectionEnd = selectionEnd; + + updateInput(target); +} + +addEventListener('keydown', (event) => { + keyupEditAttention(event); +}); diff --git a/javascript/extensions.js b/javascript/extensions.js new file mode 100644 index 0000000000000000000000000000000000000000..efeaf3a5b66f8cfa18347c07f600cd90bfe02377 --- /dev/null +++ b/javascript/extensions.js @@ -0,0 +1,74 @@ + +function extensions_apply(_disabled_list, _update_list, disable_all) { + var disable = []; + var update = []; + + gradioApp().querySelectorAll('#extensions input[type="checkbox"]').forEach(function(x) { + if (x.name.startsWith("enable_") && !x.checked) { + disable.push(x.name.substring(7)); + } + + if (x.name.startsWith("update_") && x.checked) { + update.push(x.name.substring(7)); + } + }); + + restart_reload(); + + return [JSON.stringify(disable), JSON.stringify(update), disable_all]; +} + +function extensions_check() { + var disable = []; + + gradioApp().querySelectorAll('#extensions input[type="checkbox"]').forEach(function(x) { + if (x.name.startsWith("enable_") && !x.checked) { + disable.push(x.name.substring(7)); + } + }); + + gradioApp().querySelectorAll('#extensions .extension_status').forEach(function(x) { + x.innerHTML = "Loading..."; + }); + + + var id = randomId(); + requestProgress(id, gradioApp().getElementById('extensions_installed_top'), null, function() { + + }); + + return [id, JSON.stringify(disable)]; +} + +function install_extension_from_index(button, url) { + button.disabled = "disabled"; + button.value = "Installing..."; + + var textarea = gradioApp().querySelector('#extension_to_install textarea'); + textarea.value = url; + updateInput(textarea); + + gradioApp().querySelector('#install_extension_button').click(); +} + +function config_state_confirm_restore(_, config_state_name, config_restore_type) { + if (config_state_name == "Current") { + return [false, config_state_name, config_restore_type]; + } + let restored = ""; + if (config_restore_type == "extensions") { + restored = "all saved extension versions"; + } else if (config_restore_type == "webui") { + restored = "the webui version"; + } else { + restored = "the webui version and all saved extension versions"; + } + let confirmed = confirm("Are you sure you want to restore from this state?\nThis will reset " + restored + "."); + if (confirmed) { + restart_reload(); + gradioApp().querySelectorAll('#extensions .extension_status').forEach(function(x) { + x.innerHTML = "Loading..."; + }); + } + return [confirmed, config_state_name, config_restore_type]; +} diff --git a/javascript/extraNetworks.js b/javascript/extraNetworks.js new file mode 100644 index 0000000000000000000000000000000000000000..aafe0a005f8b674d2a40b644b8115a3ee9afae11 --- /dev/null +++ b/javascript/extraNetworks.js @@ -0,0 +1,215 @@ +function setupExtraNetworksForTab(tabname) { + gradioApp().querySelector('#' + tabname + '_extra_tabs').classList.add('extra-networks'); + + var tabs = gradioApp().querySelector('#' + tabname + '_extra_tabs > div'); + var search = gradioApp().querySelector('#' + tabname + '_extra_search textarea'); + var refresh = gradioApp().getElementById(tabname + '_extra_refresh'); + + search.classList.add('search'); + tabs.appendChild(search); + tabs.appendChild(refresh); + + var applyFilter = function() { + var searchTerm = search.value.toLowerCase(); + + gradioApp().querySelectorAll('#' + tabname + '_extra_tabs div.card').forEach(function(elem) { + var searchOnly = elem.querySelector('.search_only'); + var text = elem.querySelector('.name').textContent.toLowerCase() + " " + elem.querySelector('.search_term').textContent.toLowerCase(); + + var visible = text.indexOf(searchTerm) != -1; + + if (searchOnly && searchTerm.length < 4) { + visible = false; + } + + elem.style.display = visible ? "" : "none"; + }); + }; + + search.addEventListener("input", applyFilter); + applyFilter(); + + extraNetworksApplyFilter[tabname] = applyFilter; +} + +function applyExtraNetworkFilter(tabname) { + setTimeout(extraNetworksApplyFilter[tabname], 1); +} + +var extraNetworksApplyFilter = {}; +var activePromptTextarea = {}; + +function setupExtraNetworks() { + setupExtraNetworksForTab('txt2img'); + setupExtraNetworksForTab('img2img'); + + function registerPrompt(tabname, id) { + var textarea = gradioApp().querySelector("#" + id + " > label > textarea"); + + if (!activePromptTextarea[tabname]) { + activePromptTextarea[tabname] = textarea; + } + + textarea.addEventListener("focus", function() { + activePromptTextarea[tabname] = textarea; + }); + } + + registerPrompt('txt2img', 'txt2img_prompt'); + registerPrompt('txt2img', 'txt2img_neg_prompt'); + registerPrompt('img2img', 'img2img_prompt'); + registerPrompt('img2img', 'img2img_neg_prompt'); +} + +onUiLoaded(setupExtraNetworks); + +var re_extranet = /<([^:]+:[^:]+):[\d.]+>/; +var re_extranet_g = /\s+<([^:]+:[^:]+):[\d.]+>/g; + +function tryToRemoveExtraNetworkFromPrompt(textarea, text) { + var m = text.match(re_extranet); + var replaced = false; + var newTextareaText; + if (m) { + var partToSearch = m[1]; + newTextareaText = textarea.value.replaceAll(re_extranet_g, function(found) { + m = found.match(re_extranet); + if (m[1] == partToSearch) { + replaced = true; + return ""; + } + return found; + }); + } else { + newTextareaText = textarea.value.replaceAll(new RegExp(text, "g"), function(found) { + if (found == text) { + replaced = true; + return ""; + } + return found; + }); + } + + if (replaced) { + textarea.value = newTextareaText; + return true; + } + + return false; +} + +function cardClicked(tabname, textToAdd, allowNegativePrompt) { + var textarea = allowNegativePrompt ? activePromptTextarea[tabname] : gradioApp().querySelector("#" + tabname + "_prompt > label > textarea"); + + if (!tryToRemoveExtraNetworkFromPrompt(textarea, textToAdd)) { + textarea.value = textarea.value + opts.extra_networks_add_text_separator + textToAdd; + } + + updateInput(textarea); +} + +function saveCardPreview(event, tabname, filename) { + var textarea = gradioApp().querySelector("#" + tabname + '_preview_filename > label > textarea'); + var button = gradioApp().getElementById(tabname + '_save_preview'); + + textarea.value = filename; + updateInput(textarea); + + button.click(); + + event.stopPropagation(); + event.preventDefault(); +} + +function extraNetworksSearchButton(tabs_id, event) { + var searchTextarea = gradioApp().querySelector("#" + tabs_id + ' > div > textarea'); + var button = event.target; + var text = button.classList.contains("search-all") ? "" : button.textContent.trim(); + + searchTextarea.value = text; + updateInput(searchTextarea); +} + +var globalPopup = null; +var globalPopupInner = null; +function popup(contents) { + if (!globalPopup) { + globalPopup = document.createElement('div'); + globalPopup.onclick = function() { + globalPopup.style.display = "none"; + }; + globalPopup.classList.add('global-popup'); + + var close = document.createElement('div'); + close.classList.add('global-popup-close'); + close.onclick = function() { + globalPopup.style.display = "none"; + }; + close.title = "Close"; + globalPopup.appendChild(close); + + globalPopupInner = document.createElement('div'); + globalPopupInner.onclick = function(event) { + event.stopPropagation(); return false; + }; + globalPopupInner.classList.add('global-popup-inner'); + globalPopup.appendChild(globalPopupInner); + + gradioApp().appendChild(globalPopup); + } + + globalPopupInner.innerHTML = ''; + globalPopupInner.appendChild(contents); + + globalPopup.style.display = "flex"; +} + +function extraNetworksShowMetadata(text) { + var elem = document.createElement('pre'); + elem.classList.add('popup-metadata'); + elem.textContent = text; + + popup(elem); +} + +function requestGet(url, data, handler, errorHandler) { + var xhr = new XMLHttpRequest(); + var args = Object.keys(data).map(function(k) { + return encodeURIComponent(k) + '=' + encodeURIComponent(data[k]); + }).join('&'); + xhr.open("GET", url + "?" + args, true); + + xhr.onreadystatechange = function() { + if (xhr.readyState === 4) { + if (xhr.status === 200) { + try { + var js = JSON.parse(xhr.responseText); + handler(js); + } catch (error) { + console.error(error); + errorHandler(); + } + } else { + errorHandler(); + } + } + }; + var js = JSON.stringify(data); + xhr.send(js); +} + +function extraNetworksRequestMetadata(event, extraPage, cardName) { + var showError = function() { + extraNetworksShowMetadata("there was an error getting metadata"); + }; + + requestGet("./sd_extra_networks/metadata", {page: extraPage, item: cardName}, function(data) { + if (data && data.metadata) { + extraNetworksShowMetadata(data.metadata); + } else { + showError(); + } + }, showError); + + event.stopPropagation(); +} diff --git a/javascript/generationParams.js b/javascript/generationParams.js new file mode 100644 index 0000000000000000000000000000000000000000..a877f8a5474e0bf199cb0082e3f8b971bec7292d --- /dev/null +++ b/javascript/generationParams.js @@ -0,0 +1,35 @@ +// attaches listeners to the txt2img and img2img galleries to update displayed generation param text when the image changes + +let txt2img_gallery, img2img_gallery, modal = undefined; +onUiUpdate(function() { + if (!txt2img_gallery) { + txt2img_gallery = attachGalleryListeners("txt2img"); + } + if (!img2img_gallery) { + img2img_gallery = attachGalleryListeners("img2img"); + } + if (!modal) { + modal = gradioApp().getElementById('lightboxModal'); + modalObserver.observe(modal, {attributes: true, attributeFilter: ['style']}); + } +}); + +let modalObserver = new MutationObserver(function(mutations) { + mutations.forEach(function(mutationRecord) { + let selectedTab = gradioApp().querySelector('#tabs div button.selected')?.innerText; + if (mutationRecord.target.style.display === 'none' && (selectedTab === 'txt2img' || selectedTab === 'img2img')) { + gradioApp().getElementById(selectedTab + "_generation_info_button")?.click(); + } + }); +}); + +function attachGalleryListeners(tab_name) { + var gallery = gradioApp().querySelector('#' + tab_name + '_gallery'); + gallery?.addEventListener('click', () => gradioApp().getElementById(tab_name + "_generation_info_button").click()); + gallery?.addEventListener('keydown', (e) => { + if (e.keyCode == 37 || e.keyCode == 39) { // left or right arrow + gradioApp().getElementById(tab_name + "_generation_info_button").click(); + } + }); + return gallery; +} diff --git a/javascript/hints.js b/javascript/hints.js new file mode 100644 index 0000000000000000000000000000000000000000..46f342cb9cffb77059d0a3f81453806397988aad --- /dev/null +++ b/javascript/hints.js @@ -0,0 +1,168 @@ +// mouseover tooltips for various UI elements + +var titles = { + "Sampling steps": "How many times to improve the generated image iteratively; higher values take longer; very low values can produce bad results", + "Sampling method": "Which algorithm to use to produce the image", + "GFPGAN": "Restore low quality faces using GFPGAN neural network", + "Euler a": "Euler Ancestral - very creative, each can get a completely different picture depending on step count, setting steps higher than 30-40 does not help", + "DDIM": "Denoising Diffusion Implicit Models - best at inpainting", + "UniPC": "Unified Predictor-Corrector Framework for Fast Sampling of Diffusion Models", + "DPM adaptive": "Ignores step count - uses a number of steps determined by the CFG and resolution", + + "\u{1F4D0}": "Auto detect size from img2img", + "Batch count": "How many batches of images to create (has no impact on generation performance or VRAM usage)", + "Batch size": "How many image to create in a single batch (increases generation performance at cost of higher VRAM usage)", + "CFG Scale": "Classifier Free Guidance Scale - how strongly the image should conform to prompt - lower values produce more creative results", + "Seed": "A value that determines the output of random number generator - if you create an image with same parameters and seed as another image, you'll get the same result", + "\u{1f3b2}\ufe0f": "Set seed to -1, which will cause a new random number to be used every time", + "\u267b\ufe0f": "Reuse seed from last generation, mostly useful if it was randomed", + "\u2199\ufe0f": "Read generation parameters from prompt or last generation if prompt is empty into user interface.", + "\u{1f4c2}": "Open images output directory", + "\u{1f4be}": "Save style", + "\u{1f5d1}\ufe0f": "Clear prompt", + "\u{1f4cb}": "Apply selected styles to current prompt", + "\u{1f4d2}": "Paste available values into the field", + "\u{1f3b4}": "Show/hide extra networks", + "\u{1f300}": "Restore progress", + + "Inpaint a part of image": "Draw a mask over an image, and the script will regenerate the masked area with content according to prompt", + "SD upscale": "Upscale image normally, split result into tiles, improve each tile using img2img, merge whole image back", + + "Just resize": "Resize image to target resolution. Unless height and width match, you will get incorrect aspect ratio.", + "Crop and resize": "Resize the image so that entirety of target resolution is filled with the image. Crop parts that stick out.", + "Resize and fill": "Resize the image so that entirety of image is inside target resolution. Fill empty space with image's colors.", + + "Mask blur": "How much to blur the mask before processing, in pixels.", + "Masked content": "What to put inside the masked area before processing it with Stable Diffusion.", + "fill": "fill it with colors of the image", + "original": "keep whatever was there originally", + "latent noise": "fill it with latent space noise", + "latent nothing": "fill it with latent space zeroes", + "Inpaint at full resolution": "Upscale masked region to target resolution, do inpainting, downscale back and paste into original image", + + "Denoising strength": "Determines how little respect the algorithm should have for image's content. At 0, nothing will change, and at 1 you'll get an unrelated image. With values below 1.0, processing will take less steps than the Sampling Steps slider specifies.", + + "Skip": "Stop processing current image and continue processing.", + "Interrupt": "Stop processing images and return any results accumulated so far.", + "Save": "Write image to a directory (default - log/images) and generation parameters into csv file.", + + "X values": "Separate values for X axis using commas.", + "Y values": "Separate values for Y axis using commas.", + + "None": "Do not do anything special", + "Prompt matrix": "Separate prompts into parts using vertical pipe character (|) and the script will create a picture for every combination of them (except for the first part, which will be present in all combinations)", + "X/Y/Z plot": "Create grid(s) where images will have different parameters. Use inputs below to specify which parameters will be shared by columns and rows", + "Custom code": "Run Python code. Advanced user only. Must run program with --allow-code for this to work", + + "Prompt S/R": "Separate a list of words with commas, and the first word will be used as a keyword: script will search for this word in the prompt, and replace it with others", + "Prompt order": "Separate a list of words with commas, and the script will make a variation of prompt with those words for their every possible order", + + "Tiling": "Produce an image that can be tiled.", + "Tile overlap": "For SD upscale, how much overlap in pixels should there be between tiles. Tiles overlap so that when they are merged back into one picture, there is no clearly visible seam.", + + "Variation seed": "Seed of a different picture to be mixed into the generation.", + "Variation strength": "How strong of a variation to produce. At 0, there will be no effect. At 1, you will get the complete picture with variation seed (except for ancestral samplers, where you will just get something).", + "Resize seed from height": "Make an attempt to produce a picture similar to what would have been produced with same seed at specified resolution", + "Resize seed from width": "Make an attempt to produce a picture similar to what would have been produced with same seed at specified resolution", + + "Interrogate": "Reconstruct prompt from existing image and put it into the prompt field.", + + "Images filename pattern": "Use tags like [seed] and [date] to define how filenames for images are chosen. Leave empty for default.", + "Directory name pattern": "Use tags like [seed] and [date] to define how subdirectories for images and grids are chosen. Leave empty for default.", + "Max prompt words": "Set the maximum number of words to be used in the [prompt_words] option; ATTENTION: If the words are too long, they may exceed the maximum length of the file path that the system can handle", + + "Loopback": "Performs img2img processing multiple times. Output images are used as input for the next loop.", + "Loops": "How many times to process an image. Each output is used as the input of the next loop. If set to 1, behavior will be as if this script were not used.", + "Final denoising strength": "The denoising strength for the final loop of each image in the batch.", + "Denoising strength curve": "The denoising curve controls the rate of denoising strength change each loop. Aggressive: Most of the change will happen towards the start of the loops. Linear: Change will be constant through all loops. Lazy: Most of the change will happen towards the end of the loops.", + + "Style 1": "Style to apply; styles have components for both positive and negative prompts and apply to both", + "Style 2": "Style to apply; styles have components for both positive and negative prompts and apply to both", + "Apply style": "Insert selected styles into prompt fields", + "Create style": "Save current prompts as a style. If you add the token {prompt} to the text, the style uses that as a placeholder for your prompt when you use the style in the future.", + + "Checkpoint name": "Loads weights from checkpoint before making images. You can either use hash or a part of filename (as seen in settings) for checkpoint name. Recommended to use with Y axis for less switching.", + "Inpainting conditioning mask strength": "Only applies to inpainting models. Determines how strongly to mask off the original image for inpainting and img2img. 1.0 means fully masked, which is the default behaviour. 0.0 means a fully unmasked conditioning. Lower values will help preserve the overall composition of the image, but will struggle with large changes.", + + "vram": "Torch active: Peak amount of VRAM used by Torch during generation, excluding cached data.\nTorch reserved: Peak amount of VRAM allocated by Torch, including all active and cached data.\nSys VRAM: Peak amount of VRAM allocation across all applications / total GPU VRAM (peak utilization%).", + + "Eta noise seed delta": "If this values is non-zero, it will be added to seed and used to initialize RNG for noises when using samplers with Eta. You can use this to produce even more variation of images, or you can use this to match images of other software if you know what you are doing.", + + "Filename word regex": "This regular expression will be used extract words from filename, and they will be joined using the option below into label text used for training. Leave empty to keep filename text as it is.", + "Filename join string": "This string will be used to join split words into a single line if the option above is enabled.", + + "Quicksettings list": "List of setting names, separated by commas, for settings that should go to the quick access bar at the top, rather than the usual setting tab. See modules/shared.py for setting names. Requires restarting to apply.", + + "Weighted sum": "Result = A * (1 - M) + B * M", + "Add difference": "Result = A + (B - C) * M", + "No interpolation": "Result = A", + + "Initialization text": "If the number of tokens is more than the number of vectors, some may be skipped.\nLeave the textbox empty to start with zeroed out vectors", + "Learning rate": "How fast should training go. Low values will take longer to train, high values may fail to converge (not generate accurate results) and/or may break the embedding (This has happened if you see Loss: nan in the training info textbox. If this happens, you need to manually restore your embedding from an older not-broken backup).\n\nYou can set a single numeric value, or multiple learning rates using the syntax:\n\n rate_1:max_steps_1, rate_2:max_steps_2, ...\n\nEG: 0.005:100, 1e-3:1000, 1e-5\n\nWill train with rate of 0.005 for first 100 steps, then 1e-3 until 1000 steps, then 1e-5 for all remaining steps.", + + "Clip skip": "Early stopping parameter for CLIP model; 1 is stop at last layer as usual, 2 is stop at penultimate layer, etc.", + + "Approx NN": "Cheap neural network approximation. Very fast compared to VAE, but produces pictures with 4 times smaller horizontal/vertical resolution and lower quality.", + "Approx cheap": "Very cheap approximation. Very fast compared to VAE, but produces pictures with 8 times smaller horizontal/vertical resolution and extremely low quality.", + + "Hires. fix": "Use a two step process to partially create an image at smaller resolution, upscale, and then improve details in it without changing composition", + "Hires steps": "Number of sampling steps for upscaled picture. If 0, uses same as for original.", + "Upscale by": "Adjusts the size of the image by multiplying the original width and height by the selected value. Ignored if either Resize width to or Resize height to are non-zero.", + "Resize width to": "Resizes image to this width. If 0, width is inferred from either of two nearby sliders.", + "Resize height to": "Resizes image to this height. If 0, height is inferred from either of two nearby sliders.", + "Multiplier for extra networks": "When adding extra network such as Hypernetwork or Lora to prompt, use this multiplier for it.", + "Discard weights with matching name": "Regular expression; if weights's name matches it, the weights is not written to the resulting checkpoint. Use ^model_ema to discard EMA weights.", + "Extra networks tab order": "Comma-separated list of tab names; tabs listed here will appear in the extra networks UI first and in order lsited.", + "Negative Guidance minimum sigma": "Skip negative prompt for steps where image is already mostly denoised; the higher this value, the more skips there will be; provides increased performance in exchange for minor quality reduction." +}; + +function updateTooltipForSpan(span) { + if (span.title) return; // already has a title + + let tooltip = localization[titles[span.textContent]] || titles[span.textContent]; + + if (!tooltip) { + tooltip = localization[titles[span.value]] || titles[span.value]; + } + + if (!tooltip) { + for (const c of span.classList) { + if (c in titles) { + tooltip = localization[titles[c]] || titles[c]; + break; + } + } + } + + if (tooltip) { + span.title = tooltip; + } +} + +function updateTooltipForSelect(select) { + if (select.onchange != null) return; + + select.onchange = function() { + select.title = localization[titles[select.value]] || titles[select.value] || ""; + }; +} + +var observedTooltipElements = {SPAN: 1, BUTTON: 1, SELECT: 1, P: 1}; + +onUiUpdate(function(m) { + m.forEach(function(record) { + record.addedNodes.forEach(function(node) { + if (observedTooltipElements[node.tagName]) { + updateTooltipForSpan(node); + } + if (node.tagName == "SELECT") { + updateTooltipForSelect(node); + } + + if (node.querySelectorAll) { + node.querySelectorAll('span, button, select, p').forEach(updateTooltipForSpan); + node.querySelectorAll('select').forEach(updateTooltipForSelect); + } + }); + }); +}); diff --git a/javascript/hires_fix.js b/javascript/hires_fix.js new file mode 100644 index 0000000000000000000000000000000000000000..0d04ab3b424338634af3e71a2f9d8796a5f00224 --- /dev/null +++ b/javascript/hires_fix.js @@ -0,0 +1,18 @@ + +function onCalcResolutionHires(enable, width, height, hr_scale, hr_resize_x, hr_resize_y) { + function setInactive(elem, inactive) { + elem.classList.toggle('inactive', !!inactive); + } + + var hrUpscaleBy = gradioApp().getElementById('txt2img_hr_scale'); + var hrResizeX = gradioApp().getElementById('txt2img_hr_resize_x'); + var hrResizeY = gradioApp().getElementById('txt2img_hr_resize_y'); + + gradioApp().getElementById('txt2img_hires_fix_row2').style.display = opts.use_old_hires_fix_width_height ? "none" : ""; + + setInactive(hrUpscaleBy, opts.use_old_hires_fix_width_height || hr_resize_x > 0 || hr_resize_y > 0); + setInactive(hrResizeX, opts.use_old_hires_fix_width_height || hr_resize_x == 0); + setInactive(hrResizeY, opts.use_old_hires_fix_width_height || hr_resize_y == 0); + + return [enable, width, height, hr_scale, hr_resize_x, hr_resize_y]; +} diff --git a/javascript/imageMaskFix.js b/javascript/imageMaskFix.js new file mode 100644 index 0000000000000000000000000000000000000000..3c9b8a6fd6a563d9480619ffca56e0b174d3a39b --- /dev/null +++ b/javascript/imageMaskFix.js @@ -0,0 +1,43 @@ +/** + * temporary fix for https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/668 + * @see https://github.com/gradio-app/gradio/issues/1721 + */ +function imageMaskResize() { + const canvases = gradioApp().querySelectorAll('#img2maskimg .touch-none canvas'); + if (!canvases.length) { + window.removeEventListener('resize', imageMaskResize); + return; + } + + const wrapper = canvases[0].closest('.touch-none'); + const previewImage = wrapper.previousElementSibling; + + if (!previewImage.complete) { + previewImage.addEventListener('load', imageMaskResize); + return; + } + + const w = previewImage.width; + const h = previewImage.height; + const nw = previewImage.naturalWidth; + const nh = previewImage.naturalHeight; + const portrait = nh > nw; + + const wW = Math.min(w, portrait ? h / nh * nw : w / nw * nw); + const wH = Math.min(h, portrait ? h / nh * nh : w / nw * nh); + + wrapper.style.width = `${wW}px`; + wrapper.style.height = `${wH}px`; + wrapper.style.left = `0px`; + wrapper.style.top = `0px`; + + canvases.forEach(c => { + c.style.width = c.style.height = ''; + c.style.maxWidth = '100%'; + c.style.maxHeight = '100%'; + c.style.objectFit = 'contain'; + }); +} + +onUiUpdate(imageMaskResize); +window.addEventListener('resize', imageMaskResize); diff --git a/javascript/imageParams.js b/javascript/imageParams.js new file mode 100644 index 0000000000000000000000000000000000000000..057e2d3924ee627c33e7adc0c6098a0af382fc85 --- /dev/null +++ b/javascript/imageParams.js @@ -0,0 +1,18 @@ +window.onload = (function() { + window.addEventListener('drop', e => { + const target = e.composedPath()[0]; + if (target.placeholder.indexOf("Prompt") == -1) return; + + let prompt_target = get_tab_index('tabs') == 1 ? "img2img_prompt_image" : "txt2img_prompt_image"; + + e.stopPropagation(); + e.preventDefault(); + const imgParent = gradioApp().getElementById(prompt_target); + const files = e.dataTransfer.files; + const fileInput = imgParent.querySelector('input[type="file"]'); + if (fileInput) { + fileInput.files = files; + fileInput.dispatchEvent(new Event('change')); + } + }); +}); diff --git a/javascript/imageviewer.js b/javascript/imageviewer.js new file mode 100644 index 0000000000000000000000000000000000000000..78e24eb9e81561710e8092087ef761ffd6d8251a --- /dev/null +++ b/javascript/imageviewer.js @@ -0,0 +1,254 @@ +// A full size 'lightbox' preview modal shown when left clicking on gallery previews +function closeModal() { + gradioApp().getElementById("lightboxModal").style.display = "none"; +} + +function showModal(event) { + const source = event.target || event.srcElement; + const modalImage = gradioApp().getElementById("modalImage"); + const lb = gradioApp().getElementById("lightboxModal"); + modalImage.src = source.src; + if (modalImage.style.display === 'none') { + lb.style.setProperty('background-image', 'url(' + source.src + ')'); + } + lb.style.display = "flex"; + lb.focus(); + + const tabTxt2Img = gradioApp().getElementById("tab_txt2img"); + const tabImg2Img = gradioApp().getElementById("tab_img2img"); + // show the save button in modal only on txt2img or img2img tabs + if (tabTxt2Img.style.display != "none" || tabImg2Img.style.display != "none") { + gradioApp().getElementById("modal_save").style.display = "inline"; + } else { + gradioApp().getElementById("modal_save").style.display = "none"; + } + event.stopPropagation(); +} + +function negmod(n, m) { + return ((n % m) + m) % m; +} + +function updateOnBackgroundChange() { + const modalImage = gradioApp().getElementById("modalImage"); + if (modalImage && modalImage.offsetParent) { + let currentButton = selected_gallery_button(); + + if (currentButton?.children?.length > 0 && modalImage.src != currentButton.children[0].src) { + modalImage.src = currentButton.children[0].src; + if (modalImage.style.display === 'none') { + const modal = gradioApp().getElementById("lightboxModal"); + modal.style.setProperty('background-image', `url(${modalImage.src})`); + } + } + } +} + +function modalImageSwitch(offset) { + var galleryButtons = all_gallery_buttons(); + + if (galleryButtons.length > 1) { + var currentButton = selected_gallery_button(); + + var result = -1; + galleryButtons.forEach(function(v, i) { + if (v == currentButton) { + result = i; + } + }); + + if (result != -1) { + var nextButton = galleryButtons[negmod((result + offset), galleryButtons.length)]; + nextButton.click(); + const modalImage = gradioApp().getElementById("modalImage"); + const modal = gradioApp().getElementById("lightboxModal"); + modalImage.src = nextButton.children[0].src; + if (modalImage.style.display === 'none') { + modal.style.setProperty('background-image', `url(${modalImage.src})`); + } + setTimeout(function() { + modal.focus(); + }, 10); + } + } +} + +function saveImage() { + const tabTxt2Img = gradioApp().getElementById("tab_txt2img"); + const tabImg2Img = gradioApp().getElementById("tab_img2img"); + const saveTxt2Img = "save_txt2img"; + const saveImg2Img = "save_img2img"; + if (tabTxt2Img.style.display != "none") { + gradioApp().getElementById(saveTxt2Img).click(); + } else if (tabImg2Img.style.display != "none") { + gradioApp().getElementById(saveImg2Img).click(); + } else { + console.error("missing implementation for saving modal of this type"); + } +} + +function modalSaveImage(event) { + saveImage(); + event.stopPropagation(); +} + +function modalNextImage(event) { + modalImageSwitch(1); + event.stopPropagation(); +} + +function modalPrevImage(event) { + modalImageSwitch(-1); + event.stopPropagation(); +} + +function modalKeyHandler(event) { + switch (event.key) { + case "s": + saveImage(); + break; + case "ArrowLeft": + modalPrevImage(event); + break; + case "ArrowRight": + modalNextImage(event); + break; + case "Escape": + closeModal(); + break; + } +} + +function setupImageForLightbox(e) { + if (e.dataset.modded) { + return; + } + + e.dataset.modded = true; + e.style.cursor = 'pointer'; + e.style.userSelect = 'none'; + + var isFirefox = navigator.userAgent.toLowerCase().indexOf('firefox') > -1; + + // For Firefox, listening on click first switched to next image then shows the lightbox. + // If you know how to fix this without switching to mousedown event, please. + // For other browsers the event is click to make it possiblr to drag picture. + var event = isFirefox ? 'mousedown' : 'click'; + + e.addEventListener(event, function(evt) { + if (!opts.js_modal_lightbox || evt.button != 0) return; + + modalZoomSet(gradioApp().getElementById('modalImage'), opts.js_modal_lightbox_initially_zoomed); + evt.preventDefault(); + showModal(evt); + }, true); + +} + +function modalZoomSet(modalImage, enable) { + if (modalImage) modalImage.classList.toggle('modalImageFullscreen', !!enable); +} + +function modalZoomToggle(event) { + var modalImage = gradioApp().getElementById("modalImage"); + modalZoomSet(modalImage, !modalImage.classList.contains('modalImageFullscreen')); + event.stopPropagation(); +} + +function modalTileImageToggle(event) { + const modalImage = gradioApp().getElementById("modalImage"); + const modal = gradioApp().getElementById("lightboxModal"); + const isTiling = modalImage.style.display === 'none'; + if (isTiling) { + modalImage.style.display = 'block'; + modal.style.setProperty('background-image', 'none'); + } else { + modalImage.style.display = 'none'; + modal.style.setProperty('background-image', `url(${modalImage.src})`); + } + + event.stopPropagation(); +} + +onUiUpdate(function() { + var fullImg_preview = gradioApp().querySelectorAll('.gradio-gallery > div > img'); + if (fullImg_preview != null) { + fullImg_preview.forEach(setupImageForLightbox); + } + updateOnBackgroundChange(); +}); + +document.addEventListener("DOMContentLoaded", function() { + //const modalFragment = document.createDocumentFragment(); + const modal = document.createElement('div'); + modal.onclick = closeModal; + modal.id = "lightboxModal"; + modal.tabIndex = 0; + modal.addEventListener('keydown', modalKeyHandler, true); + + const modalControls = document.createElement('div'); + modalControls.className = 'modalControls gradio-container'; + modal.append(modalControls); + + const modalZoom = document.createElement('span'); + modalZoom.className = 'modalZoom cursor'; + modalZoom.innerHTML = '⤡'; + modalZoom.addEventListener('click', modalZoomToggle, true); + modalZoom.title = "Toggle zoomed view"; + modalControls.appendChild(modalZoom); + + const modalTileImage = document.createElement('span'); + modalTileImage.className = 'modalTileImage cursor'; + modalTileImage.innerHTML = '⊞'; + modalTileImage.addEventListener('click', modalTileImageToggle, true); + modalTileImage.title = "Preview tiling"; + modalControls.appendChild(modalTileImage); + + const modalSave = document.createElement("span"); + modalSave.className = "modalSave cursor"; + modalSave.id = "modal_save"; + modalSave.innerHTML = "🖫"; + modalSave.addEventListener("click", modalSaveImage, true); + modalSave.title = "Save Image(s)"; + modalControls.appendChild(modalSave); + + const modalClose = document.createElement('span'); + modalClose.className = 'modalClose cursor'; + modalClose.innerHTML = '×'; + modalClose.onclick = closeModal; + modalClose.title = "Close image viewer"; + modalControls.appendChild(modalClose); + + const modalImage = document.createElement('img'); + modalImage.id = 'modalImage'; + modalImage.onclick = closeModal; + modalImage.tabIndex = 0; + modalImage.addEventListener('keydown', modalKeyHandler, true); + modal.appendChild(modalImage); + + const modalPrev = document.createElement('a'); + modalPrev.className = 'modalPrev'; + modalPrev.innerHTML = '❮'; + modalPrev.tabIndex = 0; + modalPrev.addEventListener('click', modalPrevImage, true); + modalPrev.addEventListener('keydown', modalKeyHandler, true); + modal.appendChild(modalPrev); + + const modalNext = document.createElement('a'); + modalNext.className = 'modalNext'; + modalNext.innerHTML = '❯'; + modalNext.tabIndex = 0; + modalNext.addEventListener('click', modalNextImage, true); + modalNext.addEventListener('keydown', modalKeyHandler, true); + + modal.appendChild(modalNext); + + try { + gradioApp().appendChild(modal); + } catch (e) { + gradioApp().body.appendChild(modal); + } + + document.body.appendChild(modal); + +}); diff --git a/javascript/imageviewerGamepad.js b/javascript/imageviewerGamepad.js new file mode 100644 index 0000000000000000000000000000000000000000..31d226dee7487ef55a57b7b8eaabcddeaa856814 --- /dev/null +++ b/javascript/imageviewerGamepad.js @@ -0,0 +1,57 @@ +window.addEventListener('gamepadconnected', (e) => { + const index = e.gamepad.index; + let isWaiting = false; + setInterval(async() => { + if (!opts.js_modal_lightbox_gamepad || isWaiting) return; + const gamepad = navigator.getGamepads()[index]; + const xValue = gamepad.axes[0]; + if (xValue <= -0.3) { + modalPrevImage(e); + isWaiting = true; + } else if (xValue >= 0.3) { + modalNextImage(e); + isWaiting = true; + } + if (isWaiting) { + await sleepUntil(() => { + const xValue = navigator.getGamepads()[index].axes[0]; + if (xValue < 0.3 && xValue > -0.3) { + return true; + } + }, opts.js_modal_lightbox_gamepad_repeat); + isWaiting = false; + } + }, 10); +}); + +/* +Primarily for vr controller type pointer devices. +I use the wheel event because there's currently no way to do it properly with web xr. + */ +let isScrolling = false; +window.addEventListener('wheel', (e) => { + if (!opts.js_modal_lightbox_gamepad || isScrolling) return; + isScrolling = true; + + if (e.deltaX <= -0.6) { + modalPrevImage(e); + } else if (e.deltaX >= 0.6) { + modalNextImage(e); + } + + setTimeout(() => { + isScrolling = false; + }, opts.js_modal_lightbox_gamepad_repeat); +}); + +function sleepUntil(f, timeout) { + return new Promise((resolve) => { + const timeStart = new Date(); + const wait = setInterval(function() { + if (f() || new Date() - timeStart > timeout) { + clearInterval(wait); + resolve(); + } + }, 20); + }); +} diff --git a/javascript/localization.js b/javascript/localization.js new file mode 100644 index 0000000000000000000000000000000000000000..eb22b8a7e99c4c9a0c4d6a52c3b9acefd74464ae --- /dev/null +++ b/javascript/localization.js @@ -0,0 +1,176 @@ + +// localization = {} -- the dict with translations is created by the backend + +var ignore_ids_for_localization = { + setting_sd_hypernetwork: 'OPTION', + setting_sd_model_checkpoint: 'OPTION', + modelmerger_primary_model_name: 'OPTION', + modelmerger_secondary_model_name: 'OPTION', + modelmerger_tertiary_model_name: 'OPTION', + train_embedding: 'OPTION', + train_hypernetwork: 'OPTION', + txt2img_styles: 'OPTION', + img2img_styles: 'OPTION', + setting_random_artist_categories: 'SPAN', + setting_face_restoration_model: 'SPAN', + setting_realesrgan_enabled_models: 'SPAN', + extras_upscaler_1: 'SPAN', + extras_upscaler_2: 'SPAN', +}; + +var re_num = /^[.\d]+$/; +var re_emoji = /[\p{Extended_Pictographic}\u{1F3FB}-\u{1F3FF}\u{1F9B0}-\u{1F9B3}]/u; + +var original_lines = {}; +var translated_lines = {}; + +function hasLocalization() { + return window.localization && Object.keys(window.localization).length > 0; +} + +function textNodesUnder(el) { + var n, a = [], walk = document.createTreeWalker(el, NodeFilter.SHOW_TEXT, null, false); + while ((n = walk.nextNode())) a.push(n); + return a; +} + +function canBeTranslated(node, text) { + if (!text) return false; + if (!node.parentElement) return false; + + var parentType = node.parentElement.nodeName; + if (parentType == 'SCRIPT' || parentType == 'STYLE' || parentType == 'TEXTAREA') return false; + + if (parentType == 'OPTION' || parentType == 'SPAN') { + var pnode = node; + for (var level = 0; level < 4; level++) { + pnode = pnode.parentElement; + if (!pnode) break; + + if (ignore_ids_for_localization[pnode.id] == parentType) return false; + } + } + + if (re_num.test(text)) return false; + if (re_emoji.test(text)) return false; + return true; +} + +function getTranslation(text) { + if (!text) return undefined; + + if (translated_lines[text] === undefined) { + original_lines[text] = 1; + } + + var tl = localization[text]; + if (tl !== undefined) { + translated_lines[tl] = 1; + } + + return tl; +} + +function processTextNode(node) { + var text = node.textContent.trim(); + + if (!canBeTranslated(node, text)) return; + + var tl = getTranslation(text); + if (tl !== undefined) { + node.textContent = tl; + } +} + +function processNode(node) { + if (node.nodeType == 3) { + processTextNode(node); + return; + } + + if (node.title) { + let tl = getTranslation(node.title); + if (tl !== undefined) { + node.title = tl; + } + } + + if (node.placeholder) { + let tl = getTranslation(node.placeholder); + if (tl !== undefined) { + node.placeholder = tl; + } + } + + textNodesUnder(node).forEach(function(node) { + processTextNode(node); + }); +} + +function dumpTranslations() { + if (!hasLocalization()) { + // If we don't have any localization, + // we will not have traversed the app to find + // original_lines, so do that now. + processNode(gradioApp()); + } + var dumped = {}; + if (localization.rtl) { + dumped.rtl = true; + } + + for (const text in original_lines) { + if (dumped[text] !== undefined) continue; + dumped[text] = localization[text] || text; + } + + return dumped; +} + +function download_localization() { + var text = JSON.stringify(dumpTranslations(), null, 4); + + var element = document.createElement('a'); + element.setAttribute('href', 'data:text/plain;charset=utf-8,' + encodeURIComponent(text)); + element.setAttribute('download', "localization.json"); + element.style.display = 'none'; + document.body.appendChild(element); + + element.click(); + + document.body.removeChild(element); +} + +document.addEventListener("DOMContentLoaded", function() { + if (!hasLocalization()) { + return; + } + + onUiUpdate(function(m) { + m.forEach(function(mutation) { + mutation.addedNodes.forEach(function(node) { + processNode(node); + }); + }); + }); + + processNode(gradioApp()); + + if (localization.rtl) { // if the language is from right to left, + (new MutationObserver((mutations, observer) => { // wait for the style to load + mutations.forEach(mutation => { + mutation.addedNodes.forEach(node => { + if (node.tagName === 'STYLE') { + observer.disconnect(); + + for (const x of node.sheet.rules) { // find all rtl media rules + if (Array.from(x.media || []).includes('rtl')) { + x.media.appendMedium('all'); // enable them + } + } + } + }); + }); + })).observe(gradioApp(), {childList: true}); + } +}); diff --git a/javascript/notification.js b/javascript/notification.js new file mode 100644 index 0000000000000000000000000000000000000000..a68a76f2514d12dc4d32c03ae8f9597e66e511c1 --- /dev/null +++ b/javascript/notification.js @@ -0,0 +1,49 @@ +// Monitors the gallery and sends a browser notification when the leading image is new. + +let lastHeadImg = null; + +let notificationButton = null; + +onUiUpdate(function() { + if (notificationButton == null) { + notificationButton = gradioApp().getElementById('request_notifications'); + + if (notificationButton != null) { + notificationButton.addEventListener('click', () => { + void Notification.requestPermission(); + }, true); + } + } + + const galleryPreviews = gradioApp().querySelectorAll('div[id^="tab_"][style*="display: block"] div[id$="_results"] .thumbnail-item > img'); + + if (galleryPreviews == null) return; + + const headImg = galleryPreviews[0]?.src; + + if (headImg == null || headImg == lastHeadImg) return; + + lastHeadImg = headImg; + + // play notification sound if available + gradioApp().querySelector('#audio_notification audio')?.play(); + + if (document.hasFocus()) return; + + // Multiple copies of the images are in the DOM when one is selected. Dedup with a Set to get the real number generated. + const imgs = new Set(Array.from(galleryPreviews).map(img => img.src)); + + const notification = new Notification( + 'Stable Diffusion', + { + body: `Generated ${imgs.size > 1 ? imgs.size - opts.return_grid : 1} image${imgs.size > 1 ? 's' : ''}`, + icon: headImg, + image: headImg, + } + ); + + notification.onclick = function(_) { + parent.focus(); + this.close(); + }; +}); diff --git a/javascript/progressbar.js b/javascript/progressbar.js new file mode 100644 index 0000000000000000000000000000000000000000..29299787e30eef0c6d411dd018561ad7976ca512 --- /dev/null +++ b/javascript/progressbar.js @@ -0,0 +1,177 @@ +// code related to showing and updating progressbar shown as the image is being made + +function rememberGallerySelection() { + +} + +function getGallerySelectedIndex() { + +} + +function request(url, data, handler, errorHandler) { + var xhr = new XMLHttpRequest(); + xhr.open("POST", url, true); + xhr.setRequestHeader("Content-Type", "application/json"); + xhr.onreadystatechange = function() { + if (xhr.readyState === 4) { + if (xhr.status === 200) { + try { + var js = JSON.parse(xhr.responseText); + handler(js); + } catch (error) { + console.error(error); + errorHandler(); + } + } else { + errorHandler(); + } + } + }; + var js = JSON.stringify(data); + xhr.send(js); +} + +function pad2(x) { + return x < 10 ? '0' + x : x; +} + +function formatTime(secs) { + if (secs > 3600) { + return pad2(Math.floor(secs / 60 / 60)) + ":" + pad2(Math.floor(secs / 60) % 60) + ":" + pad2(Math.floor(secs) % 60); + } else if (secs > 60) { + return pad2(Math.floor(secs / 60)) + ":" + pad2(Math.floor(secs) % 60); + } else { + return Math.floor(secs) + "s"; + } +} + +function setTitle(progress) { + var title = 'Stable Diffusion'; + + if (opts.show_progress_in_title && progress) { + title = '[' + progress.trim() + '] ' + title; + } + + if (document.title != title) { + document.title = title; + } +} + + +function randomId() { + return "task(" + Math.random().toString(36).slice(2, 7) + Math.random().toString(36).slice(2, 7) + Math.random().toString(36).slice(2, 7) + ")"; +} + +// starts sending progress requests to "/internal/progress" uri, creating progressbar above progressbarContainer element and +// preview inside gallery element. Cleans up all created stuff when the task is over and calls atEnd. +// calls onProgress every time there is a progress update +function requestProgress(id_task, progressbarContainer, gallery, atEnd, onProgress, inactivityTimeout = 40) { + var dateStart = new Date(); + var wasEverActive = false; + var parentProgressbar = progressbarContainer.parentNode; + var parentGallery = gallery ? gallery.parentNode : null; + + var divProgress = document.createElement('div'); + divProgress.className = 'progressDiv'; + divProgress.style.display = opts.show_progressbar ? "block" : "none"; + var divInner = document.createElement('div'); + divInner.className = 'progress'; + + divProgress.appendChild(divInner); + parentProgressbar.insertBefore(divProgress, progressbarContainer); + + if (parentGallery) { + var livePreview = document.createElement('div'); + livePreview.className = 'livePreview'; + parentGallery.insertBefore(livePreview, gallery); + } + + var removeProgressBar = function() { + setTitle(""); + parentProgressbar.removeChild(divProgress); + if (parentGallery) parentGallery.removeChild(livePreview); + atEnd(); + }; + + var fun = function(id_task, id_live_preview) { + request("./internal/progress", {id_task: id_task, id_live_preview: id_live_preview}, function(res) { + if (res.completed) { + removeProgressBar(); + return; + } + + var rect = progressbarContainer.getBoundingClientRect(); + + if (rect.width) { + divProgress.style.width = rect.width + "px"; + } + + let progressText = ""; + + divInner.style.width = ((res.progress || 0) * 100.0) + '%'; + divInner.style.background = res.progress ? "" : "transparent"; + + if (res.progress > 0) { + progressText = ((res.progress || 0) * 100.0).toFixed(0) + '%'; + } + + if (res.eta) { + progressText += " ETA: " + formatTime(res.eta); + } + + + setTitle(progressText); + + if (res.textinfo && res.textinfo.indexOf("\n") == -1) { + progressText = res.textinfo + " " + progressText; + } + + divInner.textContent = progressText; + + var elapsedFromStart = (new Date() - dateStart) / 1000; + + if (res.active) wasEverActive = true; + + if (!res.active && wasEverActive) { + removeProgressBar(); + return; + } + + if (elapsedFromStart > inactivityTimeout && !res.queued && !res.active) { + removeProgressBar(); + return; + } + + + if (res.live_preview && gallery) { + rect = gallery.getBoundingClientRect(); + if (rect.width) { + livePreview.style.width = rect.width + "px"; + livePreview.style.height = rect.height + "px"; + } + + var img = new Image(); + img.onload = function() { + livePreview.appendChild(img); + if (livePreview.childElementCount > 2) { + livePreview.removeChild(livePreview.firstElementChild); + } + }; + img.src = res.live_preview; + } + + + if (onProgress) { + onProgress(res); + } + + setTimeout(() => { + fun(id_task, res.id_live_preview); + }, opts.live_preview_refresh_period || 500); + }, function() { + removeProgressBar(); + }); + }; + + fun(id_task, 0); +} diff --git a/javascript/textualInversion.js b/javascript/textualInversion.js new file mode 100644 index 0000000000000000000000000000000000000000..20443fcca01bbba6712e40136c57dbcdb78ca945 --- /dev/null +++ b/javascript/textualInversion.js @@ -0,0 +1,17 @@ + + + +function start_training_textual_inversion() { + gradioApp().querySelector('#ti_error').innerHTML = ''; + + var id = randomId(); + requestProgress(id, gradioApp().getElementById('ti_output'), gradioApp().getElementById('ti_gallery'), function() {}, function(progress) { + gradioApp().getElementById('ti_progress').innerHTML = progress.textinfo; + }); + + var res = Array.from(arguments); + + res[0] = id; + + return res; +} diff --git a/javascript/ui.js b/javascript/ui.js new file mode 100644 index 0000000000000000000000000000000000000000..648a5290ecb5a9ba311e36aa80922048fa0ffa16 --- /dev/null +++ b/javascript/ui.js @@ -0,0 +1,456 @@ +// various functions for interaction with ui.py not large enough to warrant putting them in separate files + +function set_theme(theme) { + var gradioURL = window.location.href; + if (!gradioURL.includes('?__theme=')) { + window.location.replace(gradioURL + '?__theme=' + theme); + } +} + +function all_gallery_buttons() { + var allGalleryButtons = gradioApp().querySelectorAll('[style="display: block;"].tabitem div[id$=_gallery].gradio-gallery .thumbnails > .thumbnail-item.thumbnail-small'); + var visibleGalleryButtons = []; + allGalleryButtons.forEach(function(elem) { + if (elem.parentElement.offsetParent) { + visibleGalleryButtons.push(elem); + } + }); + return visibleGalleryButtons; +} + +function selected_gallery_button() { + var allCurrentButtons = gradioApp().querySelectorAll('[style="display: block;"].tabitem div[id$=_gallery].gradio-gallery .thumbnail-item.thumbnail-small.selected'); + var visibleCurrentButton = null; + allCurrentButtons.forEach(function(elem) { + if (elem.parentElement.offsetParent) { + visibleCurrentButton = elem; + } + }); + return visibleCurrentButton; +} + +function selected_gallery_index() { + var buttons = all_gallery_buttons(); + var button = selected_gallery_button(); + + var result = -1; + buttons.forEach(function(v, i) { + if (v == button) { + result = i; + } + }); + + return result; +} + +function extract_image_from_gallery(gallery) { + if (gallery.length == 0) { + return [null]; + } + if (gallery.length == 1) { + return [gallery[0]]; + } + + var index = selected_gallery_index(); + + if (index < 0 || index >= gallery.length) { + // Use the first image in the gallery as the default + index = 0; + } + + return [gallery[index]]; +} + +window.args_to_array = Array.from; // Compatibility with e.g. extensions that may expect this to be around + +function switch_to_txt2img() { + gradioApp().querySelector('#tabs').querySelectorAll('button')[0].click(); + + return Array.from(arguments); +} + +function switch_to_img2img_tab(no) { + gradioApp().querySelector('#tabs').querySelectorAll('button')[1].click(); + gradioApp().getElementById('mode_img2img').querySelectorAll('button')[no].click(); +} +function switch_to_img2img() { + switch_to_img2img_tab(0); + return Array.from(arguments); +} + +function switch_to_sketch() { + switch_to_img2img_tab(1); + return Array.from(arguments); +} + +function switch_to_inpaint() { + switch_to_img2img_tab(2); + return Array.from(arguments); +} + +function switch_to_inpaint_sketch() { + switch_to_img2img_tab(3); + return Array.from(arguments); +} + +function switch_to_extras() { + gradioApp().querySelector('#tabs').querySelectorAll('button')[2].click(); + + return Array.from(arguments); +} + +function get_tab_index(tabId) { + let buttons = gradioApp().getElementById(tabId).querySelector('div').querySelectorAll('button'); + for (let i = 0; i < buttons.length; i++) { + if (buttons[i].classList.contains('selected')) { + return i; + } + } + return 0; +} + +function create_tab_index_args(tabId, args) { + var res = Array.from(args); + res[0] = get_tab_index(tabId); + return res; +} + +function get_img2img_tab_index() { + let res = Array.from(arguments); + res.splice(-2); + res[0] = get_tab_index('mode_img2img'); + return res; +} + +function create_submit_args(args) { + var res = Array.from(args); + + // As it is currently, txt2img and img2img send back the previous output args (txt2img_gallery, generation_info, html_info) whenever you generate a new image. + // This can lead to uploading a huge gallery of previously generated images, which leads to an unnecessary delay between submitting and beginning to generate. + // I don't know why gradio is sending outputs along with inputs, but we can prevent sending the image gallery here, which seems to be an issue for some. + // If gradio at some point stops sending outputs, this may break something + if (Array.isArray(res[res.length - 3])) { + res[res.length - 3] = null; + } + + return res; +} + +function showSubmitButtons(tabname, show) { + gradioApp().getElementById(tabname + '_interrupt').style.display = show ? "none" : "block"; + gradioApp().getElementById(tabname + '_skip').style.display = show ? "none" : "block"; +} + +function showRestoreProgressButton(tabname, show) { + var button = gradioApp().getElementById(tabname + "_restore_progress"); + if (!button) return; + + button.style.display = show ? "flex" : "none"; +} + +function submit() { + showSubmitButtons('txt2img', false); + + var id = randomId(); + localStorage.setItem("txt2img_task_id", id); + + requestProgress(id, gradioApp().getElementById('txt2img_gallery_container'), gradioApp().getElementById('txt2img_gallery'), function() { + showSubmitButtons('txt2img', true); + localStorage.removeItem("txt2img_task_id"); + showRestoreProgressButton('txt2img', false); + }); + + var res = create_submit_args(arguments); + + res[0] = id; + + return res; +} + +function submit_img2img() { + showSubmitButtons('img2img', false); + + var id = randomId(); + localStorage.setItem("img2img_task_id", id); + + requestProgress(id, gradioApp().getElementById('img2img_gallery_container'), gradioApp().getElementById('img2img_gallery'), function() { + showSubmitButtons('img2img', true); + localStorage.removeItem("img2img_task_id"); + showRestoreProgressButton('img2img', false); + }); + + var res = create_submit_args(arguments); + + res[0] = id; + res[1] = get_tab_index('mode_img2img'); + + return res; +} + +function restoreProgressTxt2img() { + showRestoreProgressButton("txt2img", false); + var id = localStorage.getItem("txt2img_task_id"); + + id = localStorage.getItem("txt2img_task_id"); + + if (id) { + requestProgress(id, gradioApp().getElementById('txt2img_gallery_container'), gradioApp().getElementById('txt2img_gallery'), function() { + showSubmitButtons('txt2img', true); + }, null, 0); + } + + return id; +} + +function restoreProgressImg2img() { + showRestoreProgressButton("img2img", false); + + var id = localStorage.getItem("img2img_task_id"); + + if (id) { + requestProgress(id, gradioApp().getElementById('img2img_gallery_container'), gradioApp().getElementById('img2img_gallery'), function() { + showSubmitButtons('img2img', true); + }, null, 0); + } + + return id; +} + + +onUiLoaded(function() { + showRestoreProgressButton('txt2img', localStorage.getItem("txt2img_task_id")); + showRestoreProgressButton('img2img', localStorage.getItem("img2img_task_id")); +}); + + +function modelmerger() { + var id = randomId(); + requestProgress(id, gradioApp().getElementById('modelmerger_results_panel'), null, function() {}); + + var res = create_submit_args(arguments); + res[0] = id; + return res; +} + + +function ask_for_style_name(_, prompt_text, negative_prompt_text) { + var name_ = prompt('Style name:'); + return [name_, prompt_text, negative_prompt_text]; +} + +function confirm_clear_prompt(prompt, negative_prompt) { + if (confirm("Delete prompt?")) { + prompt = ""; + negative_prompt = ""; + } + + return [prompt, negative_prompt]; +} + + +var promptTokecountUpdateFuncs = {}; + +function recalculatePromptTokens(name) { + if (promptTokecountUpdateFuncs[name]) { + promptTokecountUpdateFuncs[name](); + } +} + +function recalculate_prompts_txt2img() { + recalculatePromptTokens('txt2img_prompt'); + recalculatePromptTokens('txt2img_neg_prompt'); + return Array.from(arguments); +} + +function recalculate_prompts_img2img() { + recalculatePromptTokens('img2img_prompt'); + recalculatePromptTokens('img2img_neg_prompt'); + return Array.from(arguments); +} + + +var opts = {}; +onUiUpdate(function() { + if (Object.keys(opts).length != 0) return; + + var json_elem = gradioApp().getElementById('settings_json'); + if (json_elem == null) return; + + var textarea = json_elem.querySelector('textarea'); + var jsdata = textarea.value; + opts = JSON.parse(jsdata); + + executeCallbacks(optionsChangedCallbacks); /*global optionsChangedCallbacks*/ + + Object.defineProperty(textarea, 'value', { + set: function(newValue) { + var valueProp = Object.getOwnPropertyDescriptor(HTMLTextAreaElement.prototype, 'value'); + var oldValue = valueProp.get.call(textarea); + valueProp.set.call(textarea, newValue); + + if (oldValue != newValue) { + opts = JSON.parse(textarea.value); + } + + executeCallbacks(optionsChangedCallbacks); + }, + get: function() { + var valueProp = Object.getOwnPropertyDescriptor(HTMLTextAreaElement.prototype, 'value'); + return valueProp.get.call(textarea); + } + }); + + json_elem.parentElement.style.display = "none"; + + function registerTextarea(id, id_counter, id_button) { + var prompt = gradioApp().getElementById(id); + var counter = gradioApp().getElementById(id_counter); + var textarea = gradioApp().querySelector("#" + id + " > label > textarea"); + + if (counter.parentElement == prompt.parentElement) { + return; + } + + prompt.parentElement.insertBefore(counter, prompt); + prompt.parentElement.style.position = "relative"; + + promptTokecountUpdateFuncs[id] = function() { + update_token_counter(id_button); + }; + textarea.addEventListener("input", promptTokecountUpdateFuncs[id]); + } + + registerTextarea('txt2img_prompt', 'txt2img_token_counter', 'txt2img_token_button'); + registerTextarea('txt2img_neg_prompt', 'txt2img_negative_token_counter', 'txt2img_negative_token_button'); + registerTextarea('img2img_prompt', 'img2img_token_counter', 'img2img_token_button'); + registerTextarea('img2img_neg_prompt', 'img2img_negative_token_counter', 'img2img_negative_token_button'); + + var show_all_pages = gradioApp().getElementById('settings_show_all_pages'); + var settings_tabs = gradioApp().querySelector('#settings div'); + if (show_all_pages && settings_tabs) { + settings_tabs.appendChild(show_all_pages); + show_all_pages.onclick = function() { + gradioApp().querySelectorAll('#settings > div').forEach(function(elem) { + if (elem.id == "settings_tab_licenses") { + return; + } + + elem.style.display = "block"; + }); + }; + } +}); + +onOptionsChanged(function() { + var elem = gradioApp().getElementById('sd_checkpoint_hash'); + var sd_checkpoint_hash = opts.sd_checkpoint_hash || ""; + var shorthash = sd_checkpoint_hash.substring(0, 10); + + if (elem && elem.textContent != shorthash) { + elem.textContent = shorthash; + elem.title = sd_checkpoint_hash; + elem.href = "https://google.com/search?q=" + sd_checkpoint_hash; + } +}); + +let txt2img_textarea, img2img_textarea = undefined; +let wait_time = 800; +let token_timeouts = {}; + +function update_txt2img_tokens(...args) { + update_token_counter("txt2img_token_button"); + if (args.length == 2) { + return args[0]; + } + return args; +} + +function update_img2img_tokens(...args) { + update_token_counter( + "img2img_token_button" + ); + if (args.length == 2) { + return args[0]; + } + return args; +} + +function update_token_counter(button_id) { + if (token_timeouts[button_id]) { + clearTimeout(token_timeouts[button_id]); + } + token_timeouts[button_id] = setTimeout(() => gradioApp().getElementById(button_id)?.click(), wait_time); +} + +function restart_reload() { + document.body.innerHTML = '

Reloading...

'; + + var requestPing = function() { + requestGet("./internal/ping", {}, function(data) { + location.reload(); + }, function() { + setTimeout(requestPing, 500); + }); + }; + + setTimeout(requestPing, 2000); + + return []; +} + +// Simulate an `input` DOM event for Gradio Textbox component. Needed after you edit its contents in javascript, otherwise your edits +// will only visible on web page and not sent to python. +function updateInput(target) { + let e = new Event("input", {bubbles: true}); + Object.defineProperty(e, "target", {value: target}); + target.dispatchEvent(e); +} + + +var desiredCheckpointName = null; +function selectCheckpoint(name) { + desiredCheckpointName = name; + gradioApp().getElementById('change_checkpoint').click(); +} + +function currentImg2imgSourceResolution(w, h, scaleBy) { + var img = gradioApp().querySelector('#mode_img2img > div[style="display: block;"] img'); + return img ? [img.naturalWidth, img.naturalHeight, scaleBy] : [0, 0, scaleBy]; +} + +function updateImg2imgResizeToTextAfterChangingImage() { + // At the time this is called from gradio, the image has no yet been replaced. + // There may be a better solution, but this is simple and straightforward so I'm going with it. + + setTimeout(function() { + gradioApp().getElementById('img2img_update_resize_to').click(); + }, 500); + + return []; + +} + + + +function setRandomSeed(elem_id) { + var input = gradioApp().querySelector("#" + elem_id + " input"); + if (!input) return []; + + input.value = "-1"; + updateInput(input); + return []; +} + +function switchWidthHeight(tabname) { + var width = gradioApp().querySelector("#" + tabname + "_width input[type=number]"); + var height = gradioApp().querySelector("#" + tabname + "_height input[type=number]"); + if (!width || !height) return []; + + var tmp = width.value; + width.value = height.value; + height.value = tmp; + + updateInput(width); + updateInput(height); + return []; +} diff --git a/javascript/ui_settings_hints.js b/javascript/ui_settings_hints.js new file mode 100644 index 0000000000000000000000000000000000000000..e216852b59c2b5d8b8d1661638d032d2b46c0119 --- /dev/null +++ b/javascript/ui_settings_hints.js @@ -0,0 +1,62 @@ +// various hints and extra info for the settings tab + +var settingsHintsSetup = false; + +onOptionsChanged(function() { + if (settingsHintsSetup) return; + settingsHintsSetup = true; + + gradioApp().querySelectorAll('#settings [id^=setting_]').forEach(function(div) { + var name = div.id.substr(8); + var commentBefore = opts._comments_before[name]; + var commentAfter = opts._comments_after[name]; + + if (!commentBefore && !commentAfter) return; + + var span = null; + if (div.classList.contains('gradio-checkbox')) span = div.querySelector('label span'); + else if (div.classList.contains('gradio-checkboxgroup')) span = div.querySelector('span').firstChild; + else if (div.classList.contains('gradio-radio')) span = div.querySelector('span').firstChild; + else span = div.querySelector('label span').firstChild; + + if (!span) return; + + if (commentBefore) { + var comment = document.createElement('DIV'); + comment.className = 'settings-comment'; + comment.innerHTML = commentBefore; + span.parentElement.insertBefore(document.createTextNode('\xa0'), span); + span.parentElement.insertBefore(comment, span); + span.parentElement.insertBefore(document.createTextNode('\xa0'), span); + } + if (commentAfter) { + comment = document.createElement('DIV'); + comment.className = 'settings-comment'; + comment.innerHTML = commentAfter; + span.parentElement.insertBefore(comment, span.nextSibling); + span.parentElement.insertBefore(document.createTextNode('\xa0'), span.nextSibling); + } + }); +}); + +function settingsHintsShowQuicksettings() { + requestGet("./internal/quicksettings-hint", {}, function(data) { + var table = document.createElement('table'); + table.className = 'settings-value-table'; + + data.forEach(function(obj) { + var tr = document.createElement('tr'); + var td = document.createElement('td'); + td.textContent = obj.name; + tr.appendChild(td); + + td = document.createElement('td'); + td.textContent = obj.label; + tr.appendChild(td); + + table.appendChild(tr); + }); + + popup(table); + }); +} diff --git a/launch.py b/launch.py new file mode 100644 index 0000000000000000000000000000000000000000..7a58d4c7c69a6e36cf30e7a1d5cc81a8e9ba0de4 --- /dev/null +++ b/launch.py @@ -0,0 +1,38 @@ +from modules import launch_utils + + +args = launch_utils.args +python = launch_utils.python +git = launch_utils.git +index_url = launch_utils.index_url +dir_repos = launch_utils.dir_repos + +commit_hash = launch_utils.commit_hash +git_tag = launch_utils.git_tag + +run = launch_utils.run +is_installed = launch_utils.is_installed +repo_dir = launch_utils.repo_dir + +run_pip = launch_utils.run_pip +check_run_python = launch_utils.check_run_python +git_clone = launch_utils.git_clone +git_pull_recursive = launch_utils.git_pull_recursive +run_extension_installer = launch_utils.run_extension_installer +prepare_environment = launch_utils.prepare_environment +configure_for_tests = launch_utils.configure_for_tests +start = launch_utils.start + + +def main(): + if not args.skip_prepare_environment: + prepare_environment() + + if args.test_server: + configure_for_tests() + + start() + + +if __name__ == "__main__": + main() diff --git a/localizations/Put localization files here.txt b/localizations/Put localization files here.txt new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/models/Stable-diffusion/Put Stable Diffusion checkpoints here.txt b/models/Stable-diffusion/Put Stable Diffusion checkpoints here.txt new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/models/Stable-diffusion/arcane-diffusion-v3.civitai.info b/models/Stable-diffusion/arcane-diffusion-v3.civitai.info new file mode 100644 index 0000000000000000000000000000000000000000..a0609d58970ee89117f15b82e4307af76df028b9 --- /dev/null +++ b/models/Stable-diffusion/arcane-diffusion-v3.civitai.info @@ -0,0 +1,196 @@ +{ + "id": 25, + "modelId": 23, + "name": "V3", + "createdAt": "2022-11-04T19:45:47.336Z", + "updatedAt": "2022-12-02T03:20:35.198Z", + "trainedWords": [ + "arcane style" + ], + "baseModel": "SD 1.5", + "baseModelType": "Standard", + "earlyAccessTimeFrame": 0, + "description": "

This version uses the new train-text-encoder setting and improves the quality and edibility of the model immensely. Trained on 95 images from the show in 8000 steps.

", + "stats": { + "downloadCount": 4433, + "ratingCount": 15, + "rating": 4.87 + }, + "model": { + "name": "Arcane Diffusion", + "type": "Checkpoint", + "nsfw": false, + "poi": false + }, + "files": [ + { + "id": 1016, + "url": "https://5ac0637cfd0766c97916cefa3764fbdf.r2.cloudflarestorage.com/civitai-prod-settled/models/arcane-diffusion-v3.ckpt", + "sizeKB": 2082897, + "name": "arcaneDiffusion_v3.ckpt", + "type": "Model", + "metadata": { + "fp": "fp16", + "size": "full", + "format": "PickleTensor" + }, + "pickleScanResult": "Success", + "pickleScanMessage": "**Detected Pickle imports (3)**\n```\ntorch.HalfStorage\ncollections.OrderedDict\ntorch._utils._rebuild_tensor_v2\n```", + "virusScanResult": "Success", + "virusScanMessage": null, + "scannedAt": "2022-11-12T07:42:07.491Z", + "hashes": { + "AutoV1": "318A302E", + "AutoV2": "7DD0E6760F", + "SHA256": "7DD0E6760F5A6778BEBBBDAEDD96AA1038CA76C349194F0C1B5C28311E10A78B", + "CRC32": "AA61022E", + "BLAKE3": "E7807F427E2341566B9B85A953B5DBC3A2BB070D448DA26C163C4CCA79EF52CF" + }, + "primary": true, + "downloadUrl": "https://civitai.com/api/download/models/25" + } + ], + "images": [ + { + "url": "https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2ddd9312-b080-4c70-b8ae-72a3d6bfd400/width=450/127.jpeg", + "nsfw": "None", + "width": 512, + "height": 512, + "hash": "U9Hwum%10K9t~Ba#02V@9vs.VMNJ%eWoItso", + "type": "image", + "metadata": { + "hash": "U9Hwum%10K9t~Ba#02V@9vs.VMNJ%eWoItso", + "width": 512, + "height": 512 + }, + "meta": null + }, + { + "url": "https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/aea9398e-5f4e-439f-e4c1-562019e78700/width=450/9367.jpeg", + "nsfw": "None", + "width": 400, + "height": 550, + "hash": "UrJuceWC?^xup0kCIUbHRjjte.aeWrayj[j[", + "type": "image", + "metadata": { + "hash": "UrJuceWC?^xup0kCIUbHRjjte.aeWrayj[j[", + "width": 400, + "height": 550 + }, + "meta": null + }, + { + "url": "https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/bea98d9d-3abb-4b87-cfd3-9ed0a72a8200/width=450/9366.jpeg", + "nsfw": "None", + "width": 400, + "height": 550, + "hash": "U7B:X5~E0LT|6BTL.lt77#XS[A#l?cxuMxNH", + "type": "image", + "metadata": { + "hash": "U7B:X5~E0LT|6BTL.lt77#XS[A#l?cxuMxNH", + "width": 400, + "height": 550 + }, + "meta": null + }, + { + "url": "https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/8ab8464a-0f74-4d7a-5220-2e4e4d33d500/width=450/9365.jpeg", + "nsfw": "None", + "width": 400, + "height": 550, + "hash": "UBCZCe]*8_EL4TJBYPSNF?basrw]?dVrVDo#", + "type": "image", + "metadata": { + "hash": "UBCZCe]*8_EL4TJBYPSNF?basrw]?dVrVDo#", + "width": 400, + "height": 550 + }, + "meta": null + }, + { + "url": "https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2b04fbe6-417a-4da5-d319-2f42f9a99c00/width=450/9364.jpeg", + "nsfw": "None", + "width": 400, + "height": 550, + "hash": "UKGSNYM{0=%hXnozTKj^TIadvfRi?dfmNHoc", + "type": "image", + "metadata": { + "hash": "UKGSNYM{0=%hXnozTKj^TIadvfRi?dfmNHoc", + "width": 400, + "height": 550 + }, + "meta": null + }, + { + "url": "https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/76c2b07a-bf36-4447-8a7e-bef143a79700/width=450/9363.jpeg", + "nsfw": "None", + "width": 400, + "height": 550, + "hash": "USD0l?kCT}aKyFbIjGaJR.jZr?bcOFj[xGWB", + "type": "image", + "metadata": { + "hash": "USD0l?kCT}aKyFbIjGaJR.jZr?bcOFj[xGWB", + "width": 400, + "height": 550 + }, + "meta": null + }, + { + "url": "https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/b0f123fa-f78c-40dc-174c-1bdf9dbdc200/width=450/9362.jpeg", + "nsfw": "None", + "width": 400, + "height": 550, + "hash": "U9Ch~39[K|~A?Es:FsoI9@xa-VJ6kobZsqt7", + "type": "image", + "metadata": { + "hash": "U9Ch~39[K|~A?Es:FsoI9@xa-VJ6kobZsqt7", + "width": 400, + "height": 550 + }, + "meta": null + }, + { + "url": "https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/81701ffc-9bf2-4c61-b7d2-f9b8d08b3e00/width=450/9361.jpeg", + "nsfw": "None", + "width": 400, + "height": 550, + "hash": "U6DKP[DN00l:00pJHXx_0Lniznpd*JV?rWX9", + "type": "image", + "metadata": { + "hash": "U6DKP[DN00l:00pJHXx_0Lniznpd*JV?rWX9", + "width": 400, + "height": 550 + }, + "meta": null + }, + { + "url": "https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/49eeace1-78c9-4dd2-0ec7-4c1c5d319100/width=450/9360.jpeg", + "nsfw": "None", + "width": 400, + "height": 550, + "hash": "UWH^|}-SYRI=N1R5WAofWUWAt7s,xufiS5bc", + "type": "image", + "metadata": { + "hash": "UWH^|}-SYRI=N1R5WAofWUWAt7s,xufiS5bc", + "width": 400, + "height": 550 + }, + "meta": null + }, + { + "url": "https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2699c8de-fb47-431a-56ed-fc07a9fd7c00/width=450/9371.jpeg", + "nsfw": "None", + "width": 800, + "height": 450, + "hash": "UcG8ovX8RQoe~Bs:W.WC-CjFtPR+-oM|aLkB", + "type": "image", + "metadata": { + "hash": "UcG8ovX8RQoe~Bs:W.WC-CjFtPR+-oM|aLkB", + "width": 800, + "height": 450 + }, + "meta": null + } + ], + "downloadUrl": "https://civitai.com/api/download/models/25" +} \ No newline at end of file diff --git a/models/Stable-diffusion/arcane-diffusion-v3.ckpt b/models/Stable-diffusion/arcane-diffusion-v3.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..0f44a4be7b7ac36f26383ba49b1c49111951b383 --- /dev/null +++ b/models/Stable-diffusion/arcane-diffusion-v3.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7dd0e6760f5a6778bebbbdaedd96aa1038ca76c349194f0c1b5c28311e10a78b +size 2132856622 diff --git a/models/Stable-diffusion/arcane-diffusion-v3.preview.png b/models/Stable-diffusion/arcane-diffusion-v3.preview.png new file mode 100644 index 0000000000000000000000000000000000000000..eb8fb3cf7f26b5f665ec9e486cd54806b412e05b Binary files /dev/null and b/models/Stable-diffusion/arcane-diffusion-v3.preview.png differ diff --git a/models/VAE-approx/model.pt b/models/VAE-approx/model.pt new file mode 100644 index 0000000000000000000000000000000000000000..09c6b8f7fda5e15495c6203ca323d6573745d0af --- /dev/null +++ b/models/VAE-approx/model.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f88c9078bb2238cdd0d8864671dd33e3f42e091e41f08903f3c15e4a54a9b39 +size 213777 diff --git a/models/VAE/Put VAE here.txt b/models/VAE/Put VAE here.txt new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/models/deepbooru/Put your deepbooru release project folder here.txt b/models/deepbooru/Put your deepbooru release project folder here.txt new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/models/karlo/ViT-L-14_stats.th b/models/karlo/ViT-L-14_stats.th new file mode 100644 index 0000000000000000000000000000000000000000..a6a06e94ecaa4f2977972ff991f75db6c90403ea Binary files /dev/null and b/models/karlo/ViT-L-14_stats.th differ diff --git a/modules/Roboto-Regular.ttf b/modules/Roboto-Regular.ttf new file mode 100644 index 0000000000000000000000000000000000000000..500b1045b0c94d83d2e6798aaf1faa55a2dab6fc Binary files /dev/null and b/modules/Roboto-Regular.ttf differ diff --git a/modules/api/api.py b/modules/api/api.py new file mode 100644 index 0000000000000000000000000000000000000000..eee99bbb29cf263fd3390e4192d021fea081bed3 --- /dev/null +++ b/modules/api/api.py @@ -0,0 +1,703 @@ +import base64 +import io +import time +import datetime +import uvicorn +import gradio as gr +from threading import Lock +from io import BytesIO +from fastapi import APIRouter, Depends, FastAPI, Request, Response +from fastapi.security import HTTPBasic, HTTPBasicCredentials +from fastapi.exceptions import HTTPException +from fastapi.responses import JSONResponse +from fastapi.encoders import jsonable_encoder +from secrets import compare_digest + +import modules.shared as shared +from modules import sd_samplers, deepbooru, sd_hijack, images, scripts, ui, postprocessing +from modules.api import models +from modules.shared import opts +from modules.processing import StableDiffusionProcessingTxt2Img, StableDiffusionProcessingImg2Img, process_images +from modules.textual_inversion.textual_inversion import create_embedding, train_embedding +from modules.textual_inversion.preprocess import preprocess +from modules.hypernetworks.hypernetwork import create_hypernetwork, train_hypernetwork +from PIL import PngImagePlugin,Image +from modules.sd_models import checkpoints_list, unload_model_weights, reload_model_weights +from modules.sd_models_config import find_checkpoint_config_near_filename +from modules.realesrgan_model import get_realesrgan_models +from modules import devices +from typing import Dict, List, Any +import piexif +import piexif.helper + + +def upscaler_to_index(name: str): + try: + return [x.name.lower() for x in shared.sd_upscalers].index(name.lower()) + except Exception as e: + raise HTTPException(status_code=400, detail=f"Invalid upscaler, needs to be one of these: {' , '.join([x.name for x in shared.sd_upscalers])}") from e + + +def script_name_to_index(name, scripts): + try: + return [script.title().lower() for script in scripts].index(name.lower()) + except Exception as e: + raise HTTPException(status_code=422, detail=f"Script '{name}' not found") from e + + +def validate_sampler_name(name): + config = sd_samplers.all_samplers_map.get(name, None) + if config is None: + raise HTTPException(status_code=404, detail="Sampler not found") + + return name + + +def setUpscalers(req: dict): + reqDict = vars(req) + reqDict['extras_upscaler_1'] = reqDict.pop('upscaler_1', None) + reqDict['extras_upscaler_2'] = reqDict.pop('upscaler_2', None) + return reqDict + + +def decode_base64_to_image(encoding): + if encoding.startswith("data:image/"): + encoding = encoding.split(";")[1].split(",")[1] + try: + image = Image.open(BytesIO(base64.b64decode(encoding))) + return image + except Exception as e: + raise HTTPException(status_code=500, detail="Invalid encoded image") from e + + +def encode_pil_to_base64(image): + with io.BytesIO() as output_bytes: + + if opts.samples_format.lower() == 'png': + use_metadata = False + metadata = PngImagePlugin.PngInfo() + for key, value in image.info.items(): + if isinstance(key, str) and isinstance(value, str): + metadata.add_text(key, value) + use_metadata = True + image.save(output_bytes, format="PNG", pnginfo=(metadata if use_metadata else None), quality=opts.jpeg_quality) + + elif opts.samples_format.lower() in ("jpg", "jpeg", "webp"): + parameters = image.info.get('parameters', None) + exif_bytes = piexif.dump({ + "Exif": { piexif.ExifIFD.UserComment: piexif.helper.UserComment.dump(parameters or "", encoding="unicode") } + }) + if opts.samples_format.lower() in ("jpg", "jpeg"): + image.save(output_bytes, format="JPEG", exif = exif_bytes, quality=opts.jpeg_quality) + else: + image.save(output_bytes, format="WEBP", exif = exif_bytes, quality=opts.jpeg_quality) + + else: + raise HTTPException(status_code=500, detail="Invalid image format") + + bytes_data = output_bytes.getvalue() + + return base64.b64encode(bytes_data) + + +def api_middleware(app: FastAPI): + rich_available = True + try: + import anyio # importing just so it can be placed on silent list + import starlette # importing just so it can be placed on silent list + from rich.console import Console + console = Console() + except Exception: + import traceback + rich_available = False + + @app.middleware("http") + async def log_and_time(req: Request, call_next): + ts = time.time() + res: Response = await call_next(req) + duration = str(round(time.time() - ts, 4)) + res.headers["X-Process-Time"] = duration + endpoint = req.scope.get('path', 'err') + if shared.cmd_opts.api_log and endpoint.startswith('/sdapi'): + print('API {t} {code} {prot}/{ver} {method} {endpoint} {cli} {duration}'.format( + t = datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S.%f"), + code = res.status_code, + ver = req.scope.get('http_version', '0.0'), + cli = req.scope.get('client', ('0:0.0.0', 0))[0], + prot = req.scope.get('scheme', 'err'), + method = req.scope.get('method', 'err'), + endpoint = endpoint, + duration = duration, + )) + return res + + def handle_exception(request: Request, e: Exception): + err = { + "error": type(e).__name__, + "detail": vars(e).get('detail', ''), + "body": vars(e).get('body', ''), + "errors": str(e), + } + if not isinstance(e, HTTPException): # do not print backtrace on known httpexceptions + print(f"API error: {request.method}: {request.url} {err}") + if rich_available: + console.print_exception(show_locals=True, max_frames=2, extra_lines=1, suppress=[anyio, starlette], word_wrap=False, width=min([console.width, 200])) + else: + traceback.print_exc() + return JSONResponse(status_code=vars(e).get('status_code', 500), content=jsonable_encoder(err)) + + @app.middleware("http") + async def exception_handling(request: Request, call_next): + try: + return await call_next(request) + except Exception as e: + return handle_exception(request, e) + + @app.exception_handler(Exception) + async def fastapi_exception_handler(request: Request, e: Exception): + return handle_exception(request, e) + + @app.exception_handler(HTTPException) + async def http_exception_handler(request: Request, e: HTTPException): + return handle_exception(request, e) + + +class Api: + def __init__(self, app: FastAPI, queue_lock: Lock): + if shared.cmd_opts.api_auth: + self.credentials = {} + for auth in shared.cmd_opts.api_auth.split(","): + user, password = auth.split(":") + self.credentials[user] = password + + self.router = APIRouter() + self.app = app + self.queue_lock = queue_lock + api_middleware(self.app) + self.add_api_route("/sdapi/v1/txt2img", self.text2imgapi, methods=["POST"], response_model=models.TextToImageResponse) + self.add_api_route("/sdapi/v1/img2img", self.img2imgapi, methods=["POST"], response_model=models.ImageToImageResponse) + self.add_api_route("/sdapi/v1/extra-single-image", self.extras_single_image_api, methods=["POST"], response_model=models.ExtrasSingleImageResponse) + self.add_api_route("/sdapi/v1/extra-batch-images", self.extras_batch_images_api, methods=["POST"], response_model=models.ExtrasBatchImagesResponse) + self.add_api_route("/sdapi/v1/png-info", self.pnginfoapi, methods=["POST"], response_model=models.PNGInfoResponse) + self.add_api_route("/sdapi/v1/progress", self.progressapi, methods=["GET"], response_model=models.ProgressResponse) + self.add_api_route("/sdapi/v1/interrogate", self.interrogateapi, methods=["POST"]) + self.add_api_route("/sdapi/v1/interrupt", self.interruptapi, methods=["POST"]) + self.add_api_route("/sdapi/v1/skip", self.skip, methods=["POST"]) + self.add_api_route("/sdapi/v1/options", self.get_config, methods=["GET"], response_model=models.OptionsModel) + self.add_api_route("/sdapi/v1/options", self.set_config, methods=["POST"]) + self.add_api_route("/sdapi/v1/cmd-flags", self.get_cmd_flags, methods=["GET"], response_model=models.FlagsModel) + self.add_api_route("/sdapi/v1/samplers", self.get_samplers, methods=["GET"], response_model=List[models.SamplerItem]) + self.add_api_route("/sdapi/v1/upscalers", self.get_upscalers, methods=["GET"], response_model=List[models.UpscalerItem]) + self.add_api_route("/sdapi/v1/sd-models", self.get_sd_models, methods=["GET"], response_model=List[models.SDModelItem]) + self.add_api_route("/sdapi/v1/hypernetworks", self.get_hypernetworks, methods=["GET"], response_model=List[models.HypernetworkItem]) + self.add_api_route("/sdapi/v1/face-restorers", self.get_face_restorers, methods=["GET"], response_model=List[models.FaceRestorerItem]) + self.add_api_route("/sdapi/v1/realesrgan-models", self.get_realesrgan_models, methods=["GET"], response_model=List[models.RealesrganItem]) + self.add_api_route("/sdapi/v1/prompt-styles", self.get_prompt_styles, methods=["GET"], response_model=List[models.PromptStyleItem]) + self.add_api_route("/sdapi/v1/embeddings", self.get_embeddings, methods=["GET"], response_model=models.EmbeddingsResponse) + self.add_api_route("/sdapi/v1/refresh-checkpoints", self.refresh_checkpoints, methods=["POST"]) + self.add_api_route("/sdapi/v1/create/embedding", self.create_embedding, methods=["POST"], response_model=models.CreateResponse) + self.add_api_route("/sdapi/v1/create/hypernetwork", self.create_hypernetwork, methods=["POST"], response_model=models.CreateResponse) + self.add_api_route("/sdapi/v1/preprocess", self.preprocess, methods=["POST"], response_model=models.PreprocessResponse) + self.add_api_route("/sdapi/v1/train/embedding", self.train_embedding, methods=["POST"], response_model=models.TrainResponse) + self.add_api_route("/sdapi/v1/train/hypernetwork", self.train_hypernetwork, methods=["POST"], response_model=models.TrainResponse) + self.add_api_route("/sdapi/v1/memory", self.get_memory, methods=["GET"], response_model=models.MemoryResponse) + self.add_api_route("/sdapi/v1/unload-checkpoint", self.unloadapi, methods=["POST"]) + self.add_api_route("/sdapi/v1/reload-checkpoint", self.reloadapi, methods=["POST"]) + self.add_api_route("/sdapi/v1/scripts", self.get_scripts_list, methods=["GET"], response_model=models.ScriptsList) + self.add_api_route("/sdapi/v1/script-info", self.get_script_info, methods=["GET"], response_model=List[models.ScriptInfo]) + + self.default_script_arg_txt2img = [] + self.default_script_arg_img2img = [] + + def add_api_route(self, path: str, endpoint, **kwargs): + if shared.cmd_opts.api_auth: + return self.app.add_api_route(path, endpoint, dependencies=[Depends(self.auth)], **kwargs) + return self.app.add_api_route(path, endpoint, **kwargs) + + def auth(self, credentials: HTTPBasicCredentials = Depends(HTTPBasic())): + if credentials.username in self.credentials: + if compare_digest(credentials.password, self.credentials[credentials.username]): + return True + + raise HTTPException(status_code=401, detail="Incorrect username or password", headers={"WWW-Authenticate": "Basic"}) + + def get_selectable_script(self, script_name, script_runner): + if script_name is None or script_name == "": + return None, None + + script_idx = script_name_to_index(script_name, script_runner.selectable_scripts) + script = script_runner.selectable_scripts[script_idx] + return script, script_idx + + def get_scripts_list(self): + t2ilist = [script.name for script in scripts.scripts_txt2img.scripts if script.name is not None] + i2ilist = [script.name for script in scripts.scripts_img2img.scripts if script.name is not None] + + return models.ScriptsList(txt2img=t2ilist, img2img=i2ilist) + + def get_script_info(self): + res = [] + + for script_list in [scripts.scripts_txt2img.scripts, scripts.scripts_img2img.scripts]: + res += [script.api_info for script in script_list if script.api_info is not None] + + return res + + def get_script(self, script_name, script_runner): + if script_name is None or script_name == "": + return None, None + + script_idx = script_name_to_index(script_name, script_runner.scripts) + return script_runner.scripts[script_idx] + + def init_default_script_args(self, script_runner): + #find max idx from the scripts in runner and generate a none array to init script_args + last_arg_index = 1 + for script in script_runner.scripts: + if last_arg_index < script.args_to: + last_arg_index = script.args_to + # None everywhere except position 0 to initialize script args + script_args = [None]*last_arg_index + script_args[0] = 0 + + # get default values + with gr.Blocks(): # will throw errors calling ui function without this + for script in script_runner.scripts: + if script.ui(script.is_img2img): + ui_default_values = [] + for elem in script.ui(script.is_img2img): + ui_default_values.append(elem.value) + script_args[script.args_from:script.args_to] = ui_default_values + return script_args + + def init_script_args(self, request, default_script_args, selectable_scripts, selectable_idx, script_runner): + script_args = default_script_args.copy() + # position 0 in script_arg is the idx+1 of the selectable script that is going to be run when using scripts.scripts_*2img.run() + if selectable_scripts: + script_args[selectable_scripts.args_from:selectable_scripts.args_to] = request.script_args + script_args[0] = selectable_idx + 1 + + # Now check for always on scripts + if request.alwayson_scripts and (len(request.alwayson_scripts) > 0): + for alwayson_script_name in request.alwayson_scripts.keys(): + alwayson_script = self.get_script(alwayson_script_name, script_runner) + if alwayson_script is None: + raise HTTPException(status_code=422, detail=f"always on script {alwayson_script_name} not found") + # Selectable script in always on script param check + if alwayson_script.alwayson is False: + raise HTTPException(status_code=422, detail="Cannot have a selectable script in the always on scripts params") + # always on script with no arg should always run so you don't really need to add them to the requests + if "args" in request.alwayson_scripts[alwayson_script_name]: + # min between arg length in scriptrunner and arg length in the request + for idx in range(0, min((alwayson_script.args_to - alwayson_script.args_from), len(request.alwayson_scripts[alwayson_script_name]["args"]))): + script_args[alwayson_script.args_from + idx] = request.alwayson_scripts[alwayson_script_name]["args"][idx] + return script_args + + def text2imgapi(self, txt2imgreq: models.StableDiffusionTxt2ImgProcessingAPI): + script_runner = scripts.scripts_txt2img + if not script_runner.scripts: + script_runner.initialize_scripts(False) + ui.create_ui() + if not self.default_script_arg_txt2img: + self.default_script_arg_txt2img = self.init_default_script_args(script_runner) + selectable_scripts, selectable_script_idx = self.get_selectable_script(txt2imgreq.script_name, script_runner) + + populate = txt2imgreq.copy(update={ # Override __init__ params + "sampler_name": validate_sampler_name(txt2imgreq.sampler_name or txt2imgreq.sampler_index), + "do_not_save_samples": not txt2imgreq.save_images, + "do_not_save_grid": not txt2imgreq.save_images, + }) + if populate.sampler_name: + populate.sampler_index = None # prevent a warning later on + + args = vars(populate) + args.pop('script_name', None) + args.pop('script_args', None) # will refeed them to the pipeline directly after initializing them + args.pop('alwayson_scripts', None) + + script_args = self.init_script_args(txt2imgreq, self.default_script_arg_txt2img, selectable_scripts, selectable_script_idx, script_runner) + + send_images = args.pop('send_images', True) + args.pop('save_images', None) + + with self.queue_lock: + p = StableDiffusionProcessingTxt2Img(sd_model=shared.sd_model, **args) + p.scripts = script_runner + p.outpath_grids = opts.outdir_txt2img_grids + p.outpath_samples = opts.outdir_txt2img_samples + + shared.state.begin() + if selectable_scripts is not None: + p.script_args = script_args + processed = scripts.scripts_txt2img.run(p, *p.script_args) # Need to pass args as list here + else: + p.script_args = tuple(script_args) # Need to pass args as tuple here + processed = process_images(p) + shared.state.end() + + b64images = list(map(encode_pil_to_base64, processed.images)) if send_images else [] + + return models.TextToImageResponse(images=b64images, parameters=vars(txt2imgreq), info=processed.js()) + + def img2imgapi(self, img2imgreq: models.StableDiffusionImg2ImgProcessingAPI): + init_images = img2imgreq.init_images + if init_images is None: + raise HTTPException(status_code=404, detail="Init image not found") + + mask = img2imgreq.mask + if mask: + mask = decode_base64_to_image(mask) + + script_runner = scripts.scripts_img2img + if not script_runner.scripts: + script_runner.initialize_scripts(True) + ui.create_ui() + if not self.default_script_arg_img2img: + self.default_script_arg_img2img = self.init_default_script_args(script_runner) + selectable_scripts, selectable_script_idx = self.get_selectable_script(img2imgreq.script_name, script_runner) + + populate = img2imgreq.copy(update={ # Override __init__ params + "sampler_name": validate_sampler_name(img2imgreq.sampler_name or img2imgreq.sampler_index), + "do_not_save_samples": not img2imgreq.save_images, + "do_not_save_grid": not img2imgreq.save_images, + "mask": mask, + }) + if populate.sampler_name: + populate.sampler_index = None # prevent a warning later on + + args = vars(populate) + args.pop('include_init_images', None) # this is meant to be done by "exclude": True in model, but it's for a reason that I cannot determine. + args.pop('script_name', None) + args.pop('script_args', None) # will refeed them to the pipeline directly after initializing them + args.pop('alwayson_scripts', None) + + script_args = self.init_script_args(img2imgreq, self.default_script_arg_img2img, selectable_scripts, selectable_script_idx, script_runner) + + send_images = args.pop('send_images', True) + args.pop('save_images', None) + + with self.queue_lock: + p = StableDiffusionProcessingImg2Img(sd_model=shared.sd_model, **args) + p.init_images = [decode_base64_to_image(x) for x in init_images] + p.scripts = script_runner + p.outpath_grids = opts.outdir_img2img_grids + p.outpath_samples = opts.outdir_img2img_samples + + shared.state.begin() + if selectable_scripts is not None: + p.script_args = script_args + processed = scripts.scripts_img2img.run(p, *p.script_args) # Need to pass args as list here + else: + p.script_args = tuple(script_args) # Need to pass args as tuple here + processed = process_images(p) + shared.state.end() + + b64images = list(map(encode_pil_to_base64, processed.images)) if send_images else [] + + if not img2imgreq.include_init_images: + img2imgreq.init_images = None + img2imgreq.mask = None + + return models.ImageToImageResponse(images=b64images, parameters=vars(img2imgreq), info=processed.js()) + + def extras_single_image_api(self, req: models.ExtrasSingleImageRequest): + reqDict = setUpscalers(req) + + reqDict['image'] = decode_base64_to_image(reqDict['image']) + + with self.queue_lock: + result = postprocessing.run_extras(extras_mode=0, image_folder="", input_dir="", output_dir="", save_output=False, **reqDict) + + return models.ExtrasSingleImageResponse(image=encode_pil_to_base64(result[0][0]), html_info=result[1]) + + def extras_batch_images_api(self, req: models.ExtrasBatchImagesRequest): + reqDict = setUpscalers(req) + + image_list = reqDict.pop('imageList', []) + image_folder = [decode_base64_to_image(x.data) for x in image_list] + + with self.queue_lock: + result = postprocessing.run_extras(extras_mode=1, image_folder=image_folder, image="", input_dir="", output_dir="", save_output=False, **reqDict) + + return models.ExtrasBatchImagesResponse(images=list(map(encode_pil_to_base64, result[0])), html_info=result[1]) + + def pnginfoapi(self, req: models.PNGInfoRequest): + if(not req.image.strip()): + return models.PNGInfoResponse(info="") + + image = decode_base64_to_image(req.image.strip()) + if image is None: + return models.PNGInfoResponse(info="") + + geninfo, items = images.read_info_from_image(image) + if geninfo is None: + geninfo = "" + + items = {**{'parameters': geninfo}, **items} + + return models.PNGInfoResponse(info=geninfo, items=items) + + def progressapi(self, req: models.ProgressRequest = Depends()): + # copy from check_progress_call of ui.py + + if shared.state.job_count == 0: + return models.ProgressResponse(progress=0, eta_relative=0, state=shared.state.dict(), textinfo=shared.state.textinfo) + + # avoid dividing zero + progress = 0.01 + + if shared.state.job_count > 0: + progress += shared.state.job_no / shared.state.job_count + if shared.state.sampling_steps > 0: + progress += 1 / shared.state.job_count * shared.state.sampling_step / shared.state.sampling_steps + + time_since_start = time.time() - shared.state.time_start + eta = (time_since_start/progress) + eta_relative = eta-time_since_start + + progress = min(progress, 1) + + shared.state.set_current_image() + + current_image = None + if shared.state.current_image and not req.skip_current_image: + current_image = encode_pil_to_base64(shared.state.current_image) + + return models.ProgressResponse(progress=progress, eta_relative=eta_relative, state=shared.state.dict(), current_image=current_image, textinfo=shared.state.textinfo) + + def interrogateapi(self, interrogatereq: models.InterrogateRequest): + image_b64 = interrogatereq.image + if image_b64 is None: + raise HTTPException(status_code=404, detail="Image not found") + + img = decode_base64_to_image(image_b64) + img = img.convert('RGB') + + # Override object param + with self.queue_lock: + if interrogatereq.model == "clip": + processed = shared.interrogator.interrogate(img) + elif interrogatereq.model == "deepdanbooru": + processed = deepbooru.model.tag(img) + else: + raise HTTPException(status_code=404, detail="Model not found") + + return models.InterrogateResponse(caption=processed) + + def interruptapi(self): + shared.state.interrupt() + + return {} + + def unloadapi(self): + unload_model_weights() + + return {} + + def reloadapi(self): + reload_model_weights() + + return {} + + def skip(self): + shared.state.skip() + + def get_config(self): + options = {} + for key in shared.opts.data.keys(): + metadata = shared.opts.data_labels.get(key) + if(metadata is not None): + options.update({key: shared.opts.data.get(key, shared.opts.data_labels.get(key).default)}) + else: + options.update({key: shared.opts.data.get(key, None)}) + + return options + + def set_config(self, req: Dict[str, Any]): + for k, v in req.items(): + shared.opts.set(k, v) + + shared.opts.save(shared.config_filename) + return + + def get_cmd_flags(self): + return vars(shared.cmd_opts) + + def get_samplers(self): + return [{"name": sampler[0], "aliases":sampler[2], "options":sampler[3]} for sampler in sd_samplers.all_samplers] + + def get_upscalers(self): + return [ + { + "name": upscaler.name, + "model_name": upscaler.scaler.model_name, + "model_path": upscaler.data_path, + "model_url": None, + "scale": upscaler.scale, + } + for upscaler in shared.sd_upscalers + ] + + def get_sd_models(self): + return [{"title": x.title, "model_name": x.model_name, "hash": x.shorthash, "sha256": x.sha256, "filename": x.filename, "config": find_checkpoint_config_near_filename(x)} for x in checkpoints_list.values()] + + def get_hypernetworks(self): + return [{"name": name, "path": shared.hypernetworks[name]} for name in shared.hypernetworks] + + def get_face_restorers(self): + return [{"name":x.name(), "cmd_dir": getattr(x, "cmd_dir", None)} for x in shared.face_restorers] + + def get_realesrgan_models(self): + return [{"name":x.name,"path":x.data_path, "scale":x.scale} for x in get_realesrgan_models(None)] + + def get_prompt_styles(self): + styleList = [] + for k in shared.prompt_styles.styles: + style = shared.prompt_styles.styles[k] + styleList.append({"name":style[0], "prompt": style[1], "negative_prompt": style[2]}) + + return styleList + + def get_embeddings(self): + db = sd_hijack.model_hijack.embedding_db + + def convert_embedding(embedding): + return { + "step": embedding.step, + "sd_checkpoint": embedding.sd_checkpoint, + "sd_checkpoint_name": embedding.sd_checkpoint_name, + "shape": embedding.shape, + "vectors": embedding.vectors, + } + + def convert_embeddings(embeddings): + return {embedding.name: convert_embedding(embedding) for embedding in embeddings.values()} + + return { + "loaded": convert_embeddings(db.word_embeddings), + "skipped": convert_embeddings(db.skipped_embeddings), + } + + def refresh_checkpoints(self): + shared.refresh_checkpoints() + + def create_embedding(self, args: dict): + try: + shared.state.begin() + filename = create_embedding(**args) # create empty embedding + sd_hijack.model_hijack.embedding_db.load_textual_inversion_embeddings() # reload embeddings so new one can be immediately used + shared.state.end() + return models.CreateResponse(info=f"create embedding filename: {filename}") + except AssertionError as e: + shared.state.end() + return models.TrainResponse(info=f"create embedding error: {e}") + + def create_hypernetwork(self, args: dict): + try: + shared.state.begin() + filename = create_hypernetwork(**args) # create empty embedding + shared.state.end() + return models.CreateResponse(info=f"create hypernetwork filename: {filename}") + except AssertionError as e: + shared.state.end() + return models.TrainResponse(info=f"create hypernetwork error: {e}") + + def preprocess(self, args: dict): + try: + shared.state.begin() + preprocess(**args) # quick operation unless blip/booru interrogation is enabled + shared.state.end() + return models.PreprocessResponse(info = 'preprocess complete') + except KeyError as e: + shared.state.end() + return models.PreprocessResponse(info=f"preprocess error: invalid token: {e}") + except AssertionError as e: + shared.state.end() + return models.PreprocessResponse(info=f"preprocess error: {e}") + except FileNotFoundError as e: + shared.state.end() + return models.PreprocessResponse(info=f'preprocess error: {e}') + + def train_embedding(self, args: dict): + try: + shared.state.begin() + apply_optimizations = shared.opts.training_xattention_optimizations + error = None + filename = '' + if not apply_optimizations: + sd_hijack.undo_optimizations() + try: + embedding, filename = train_embedding(**args) # can take a long time to complete + except Exception as e: + error = e + finally: + if not apply_optimizations: + sd_hijack.apply_optimizations() + shared.state.end() + return models.TrainResponse(info=f"train embedding complete: filename: {filename} error: {error}") + except AssertionError as msg: + shared.state.end() + return models.TrainResponse(info=f"train embedding error: {msg}") + + def train_hypernetwork(self, args: dict): + try: + shared.state.begin() + shared.loaded_hypernetworks = [] + apply_optimizations = shared.opts.training_xattention_optimizations + error = None + filename = '' + if not apply_optimizations: + sd_hijack.undo_optimizations() + try: + hypernetwork, filename = train_hypernetwork(**args) + except Exception as e: + error = e + finally: + shared.sd_model.cond_stage_model.to(devices.device) + shared.sd_model.first_stage_model.to(devices.device) + if not apply_optimizations: + sd_hijack.apply_optimizations() + shared.state.end() + return models.TrainResponse(info=f"train embedding complete: filename: {filename} error: {error}") + except AssertionError: + shared.state.end() + return models.TrainResponse(info=f"train embedding error: {error}") + + def get_memory(self): + try: + import os + import psutil + process = psutil.Process(os.getpid()) + res = process.memory_info() # only rss is cross-platform guaranteed so we dont rely on other values + ram_total = 100 * res.rss / process.memory_percent() # and total memory is calculated as actual value is not cross-platform safe + ram = { 'free': ram_total - res.rss, 'used': res.rss, 'total': ram_total } + except Exception as err: + ram = { 'error': f'{err}' } + try: + import torch + if torch.cuda.is_available(): + s = torch.cuda.mem_get_info() + system = { 'free': s[0], 'used': s[1] - s[0], 'total': s[1] } + s = dict(torch.cuda.memory_stats(shared.device)) + allocated = { 'current': s['allocated_bytes.all.current'], 'peak': s['allocated_bytes.all.peak'] } + reserved = { 'current': s['reserved_bytes.all.current'], 'peak': s['reserved_bytes.all.peak'] } + active = { 'current': s['active_bytes.all.current'], 'peak': s['active_bytes.all.peak'] } + inactive = { 'current': s['inactive_split_bytes.all.current'], 'peak': s['inactive_split_bytes.all.peak'] } + warnings = { 'retries': s['num_alloc_retries'], 'oom': s['num_ooms'] } + cuda = { + 'system': system, + 'active': active, + 'allocated': allocated, + 'reserved': reserved, + 'inactive': inactive, + 'events': warnings, + } + else: + cuda = {'error': 'unavailable'} + except Exception as err: + cuda = {'error': f'{err}'} + return models.MemoryResponse(ram=ram, cuda=cuda) + + def launch(self, server_name, port): + self.app.include_router(self.router) + uvicorn.run(self.app, host=server_name, port=port) diff --git a/modules/api/models.py b/modules/api/models.py new file mode 100644 index 0000000000000000000000000000000000000000..1ff2fb338cebac1cd3d996599e0a0fce3712dd43 --- /dev/null +++ b/modules/api/models.py @@ -0,0 +1,309 @@ +import inspect +from pydantic import BaseModel, Field, create_model +from typing import Any, Optional +from typing_extensions import Literal +from inflection import underscore +from modules.processing import StableDiffusionProcessingTxt2Img, StableDiffusionProcessingImg2Img +from modules.shared import sd_upscalers, opts, parser +from typing import Dict, List + +API_NOT_ALLOWED = [ + "self", + "kwargs", + "sd_model", + "outpath_samples", + "outpath_grids", + "sampler_index", + # "do_not_save_samples", + # "do_not_save_grid", + "extra_generation_params", + "overlay_images", + "do_not_reload_embeddings", + "seed_enable_extras", + "prompt_for_display", + "sampler_noise_scheduler_override", + "ddim_discretize" +] + +class ModelDef(BaseModel): + """Assistance Class for Pydantic Dynamic Model Generation""" + + field: str + field_alias: str + field_type: Any + field_value: Any + field_exclude: bool = False + + +class PydanticModelGenerator: + """ + Takes in created classes and stubs them out in a way FastAPI/Pydantic is happy about: + source_data is a snapshot of the default values produced by the class + params are the names of the actual keys required by __init__ + """ + + def __init__( + self, + model_name: str = None, + class_instance = None, + additional_fields = None, + ): + def field_type_generator(k, v): + # field_type = str if not overrides.get(k) else overrides[k]["type"] + # print(k, v.annotation, v.default) + field_type = v.annotation + + return Optional[field_type] + + def merge_class_params(class_): + all_classes = list(filter(lambda x: x is not object, inspect.getmro(class_))) + parameters = {} + for classes in all_classes: + parameters = {**parameters, **inspect.signature(classes.__init__).parameters} + return parameters + + + self._model_name = model_name + self._class_data = merge_class_params(class_instance) + + self._model_def = [ + ModelDef( + field=underscore(k), + field_alias=k, + field_type=field_type_generator(k, v), + field_value=v.default + ) + for (k,v) in self._class_data.items() if k not in API_NOT_ALLOWED + ] + + for fields in additional_fields: + self._model_def.append(ModelDef( + field=underscore(fields["key"]), + field_alias=fields["key"], + field_type=fields["type"], + field_value=fields["default"], + field_exclude=fields["exclude"] if "exclude" in fields else False)) + + def generate_model(self): + """ + Creates a pydantic BaseModel + from the json and overrides provided at initialization + """ + fields = { + d.field: (d.field_type, Field(default=d.field_value, alias=d.field_alias, exclude=d.field_exclude)) for d in self._model_def + } + DynamicModel = create_model(self._model_name, **fields) + DynamicModel.__config__.allow_population_by_field_name = True + DynamicModel.__config__.allow_mutation = True + return DynamicModel + +StableDiffusionTxt2ImgProcessingAPI = PydanticModelGenerator( + "StableDiffusionProcessingTxt2Img", + StableDiffusionProcessingTxt2Img, + [ + {"key": "sampler_index", "type": str, "default": "Euler"}, + {"key": "script_name", "type": str, "default": None}, + {"key": "script_args", "type": list, "default": []}, + {"key": "send_images", "type": bool, "default": True}, + {"key": "save_images", "type": bool, "default": False}, + {"key": "alwayson_scripts", "type": dict, "default": {}}, + ] +).generate_model() + +StableDiffusionImg2ImgProcessingAPI = PydanticModelGenerator( + "StableDiffusionProcessingImg2Img", + StableDiffusionProcessingImg2Img, + [ + {"key": "sampler_index", "type": str, "default": "Euler"}, + {"key": "init_images", "type": list, "default": None}, + {"key": "denoising_strength", "type": float, "default": 0.75}, + {"key": "mask", "type": str, "default": None}, + {"key": "include_init_images", "type": bool, "default": False, "exclude" : True}, + {"key": "script_name", "type": str, "default": None}, + {"key": "script_args", "type": list, "default": []}, + {"key": "send_images", "type": bool, "default": True}, + {"key": "save_images", "type": bool, "default": False}, + {"key": "alwayson_scripts", "type": dict, "default": {}}, + ] +).generate_model() + +class TextToImageResponse(BaseModel): + images: List[str] = Field(default=None, title="Image", description="The generated image in base64 format.") + parameters: dict + info: str + +class ImageToImageResponse(BaseModel): + images: List[str] = Field(default=None, title="Image", description="The generated image in base64 format.") + parameters: dict + info: str + +class ExtrasBaseRequest(BaseModel): + resize_mode: Literal[0, 1] = Field(default=0, title="Resize Mode", description="Sets the resize mode: 0 to upscale by upscaling_resize amount, 1 to upscale up to upscaling_resize_h x upscaling_resize_w.") + show_extras_results: bool = Field(default=True, title="Show results", description="Should the backend return the generated image?") + gfpgan_visibility: float = Field(default=0, title="GFPGAN Visibility", ge=0, le=1, allow_inf_nan=False, description="Sets the visibility of GFPGAN, values should be between 0 and 1.") + codeformer_visibility: float = Field(default=0, title="CodeFormer Visibility", ge=0, le=1, allow_inf_nan=False, description="Sets the visibility of CodeFormer, values should be between 0 and 1.") + codeformer_weight: float = Field(default=0, title="CodeFormer Weight", ge=0, le=1, allow_inf_nan=False, description="Sets the weight of CodeFormer, values should be between 0 and 1.") + upscaling_resize: float = Field(default=2, title="Upscaling Factor", ge=1, le=8, description="By how much to upscale the image, only used when resize_mode=0.") + upscaling_resize_w: int = Field(default=512, title="Target Width", ge=1, description="Target width for the upscaler to hit. Only used when resize_mode=1.") + upscaling_resize_h: int = Field(default=512, title="Target Height", ge=1, description="Target height for the upscaler to hit. Only used when resize_mode=1.") + upscaling_crop: bool = Field(default=True, title="Crop to fit", description="Should the upscaler crop the image to fit in the chosen size?") + upscaler_1: str = Field(default="None", title="Main upscaler", description=f"The name of the main upscaler to use, it has to be one of this list: {' , '.join([x.name for x in sd_upscalers])}") + upscaler_2: str = Field(default="None", title="Secondary upscaler", description=f"The name of the secondary upscaler to use, it has to be one of this list: {' , '.join([x.name for x in sd_upscalers])}") + extras_upscaler_2_visibility: float = Field(default=0, title="Secondary upscaler visibility", ge=0, le=1, allow_inf_nan=False, description="Sets the visibility of secondary upscaler, values should be between 0 and 1.") + upscale_first: bool = Field(default=False, title="Upscale first", description="Should the upscaler run before restoring faces?") + +class ExtraBaseResponse(BaseModel): + html_info: str = Field(title="HTML info", description="A series of HTML tags containing the process info.") + +class ExtrasSingleImageRequest(ExtrasBaseRequest): + image: str = Field(default="", title="Image", description="Image to work on, must be a Base64 string containing the image's data.") + +class ExtrasSingleImageResponse(ExtraBaseResponse): + image: str = Field(default=None, title="Image", description="The generated image in base64 format.") + +class FileData(BaseModel): + data: str = Field(title="File data", description="Base64 representation of the file") + name: str = Field(title="File name") + +class ExtrasBatchImagesRequest(ExtrasBaseRequest): + imageList: List[FileData] = Field(title="Images", description="List of images to work on. Must be Base64 strings") + +class ExtrasBatchImagesResponse(ExtraBaseResponse): + images: List[str] = Field(title="Images", description="The generated images in base64 format.") + +class PNGInfoRequest(BaseModel): + image: str = Field(title="Image", description="The base64 encoded PNG image") + +class PNGInfoResponse(BaseModel): + info: str = Field(title="Image info", description="A string with the parameters used to generate the image") + items: dict = Field(title="Items", description="An object containing all the info the image had") + +class ProgressRequest(BaseModel): + skip_current_image: bool = Field(default=False, title="Skip current image", description="Skip current image serialization") + +class ProgressResponse(BaseModel): + progress: float = Field(title="Progress", description="The progress with a range of 0 to 1") + eta_relative: float = Field(title="ETA in secs") + state: dict = Field(title="State", description="The current state snapshot") + current_image: str = Field(default=None, title="Current image", description="The current image in base64 format. opts.show_progress_every_n_steps is required for this to work.") + textinfo: str = Field(default=None, title="Info text", description="Info text used by WebUI.") + +class InterrogateRequest(BaseModel): + image: str = Field(default="", title="Image", description="Image to work on, must be a Base64 string containing the image's data.") + model: str = Field(default="clip", title="Model", description="The interrogate model used.") + +class InterrogateResponse(BaseModel): + caption: str = Field(default=None, title="Caption", description="The generated caption for the image.") + +class TrainResponse(BaseModel): + info: str = Field(title="Train info", description="Response string from train embedding or hypernetwork task.") + +class CreateResponse(BaseModel): + info: str = Field(title="Create info", description="Response string from create embedding or hypernetwork task.") + +class PreprocessResponse(BaseModel): + info: str = Field(title="Preprocess info", description="Response string from preprocessing task.") + +fields = {} +for key, metadata in opts.data_labels.items(): + value = opts.data.get(key) + optType = opts.typemap.get(type(metadata.default), type(value)) + + if (metadata is not None): + fields.update({key: (Optional[optType], Field( + default=metadata.default ,description=metadata.label))}) + else: + fields.update({key: (Optional[optType], Field())}) + +OptionsModel = create_model("Options", **fields) + +flags = {} +_options = vars(parser)['_option_string_actions'] +for key in _options: + if(_options[key].dest != 'help'): + flag = _options[key] + _type = str + if _options[key].default is not None: + _type = type(_options[key].default) + flags.update({flag.dest: (_type, Field(default=flag.default, description=flag.help))}) + +FlagsModel = create_model("Flags", **flags) + +class SamplerItem(BaseModel): + name: str = Field(title="Name") + aliases: List[str] = Field(title="Aliases") + options: Dict[str, str] = Field(title="Options") + +class UpscalerItem(BaseModel): + name: str = Field(title="Name") + model_name: Optional[str] = Field(title="Model Name") + model_path: Optional[str] = Field(title="Path") + model_url: Optional[str] = Field(title="URL") + scale: Optional[float] = Field(title="Scale") + +class SDModelItem(BaseModel): + title: str = Field(title="Title") + model_name: str = Field(title="Model Name") + hash: Optional[str] = Field(title="Short hash") + sha256: Optional[str] = Field(title="sha256 hash") + filename: str = Field(title="Filename") + config: Optional[str] = Field(title="Config file") + +class HypernetworkItem(BaseModel): + name: str = Field(title="Name") + path: Optional[str] = Field(title="Path") + +class FaceRestorerItem(BaseModel): + name: str = Field(title="Name") + cmd_dir: Optional[str] = Field(title="Path") + +class RealesrganItem(BaseModel): + name: str = Field(title="Name") + path: Optional[str] = Field(title="Path") + scale: Optional[int] = Field(title="Scale") + +class PromptStyleItem(BaseModel): + name: str = Field(title="Name") + prompt: Optional[str] = Field(title="Prompt") + negative_prompt: Optional[str] = Field(title="Negative Prompt") + +class ArtistItem(BaseModel): + name: str = Field(title="Name") + score: float = Field(title="Score") + category: str = Field(title="Category") + +class EmbeddingItem(BaseModel): + step: Optional[int] = Field(title="Step", description="The number of steps that were used to train this embedding, if available") + sd_checkpoint: Optional[str] = Field(title="SD Checkpoint", description="The hash of the checkpoint this embedding was trained on, if available") + sd_checkpoint_name: Optional[str] = Field(title="SD Checkpoint Name", description="The name of the checkpoint this embedding was trained on, if available. Note that this is the name that was used by the trainer; for a stable identifier, use `sd_checkpoint` instead") + shape: int = Field(title="Shape", description="The length of each individual vector in the embedding") + vectors: int = Field(title="Vectors", description="The number of vectors in the embedding") + +class EmbeddingsResponse(BaseModel): + loaded: Dict[str, EmbeddingItem] = Field(title="Loaded", description="Embeddings loaded for the current model") + skipped: Dict[str, EmbeddingItem] = Field(title="Skipped", description="Embeddings skipped for the current model (likely due to architecture incompatibility)") + +class MemoryResponse(BaseModel): + ram: dict = Field(title="RAM", description="System memory stats") + cuda: dict = Field(title="CUDA", description="nVidia CUDA memory stats") + + +class ScriptsList(BaseModel): + txt2img: list = Field(default=None, title="Txt2img", description="Titles of scripts (txt2img)") + img2img: list = Field(default=None, title="Img2img", description="Titles of scripts (img2img)") + + +class ScriptArg(BaseModel): + label: str = Field(default=None, title="Label", description="Name of the argument in UI") + value: Optional[Any] = Field(default=None, title="Value", description="Default value of the argument") + minimum: Optional[Any] = Field(default=None, title="Minimum", description="Minimum allowed value for the argumentin UI") + maximum: Optional[Any] = Field(default=None, title="Minimum", description="Maximum allowed value for the argumentin UI") + step: Optional[Any] = Field(default=None, title="Minimum", description="Step for changing value of the argumentin UI") + choices: Optional[List[str]] = Field(default=None, title="Choices", description="Possible values for the argument") + + +class ScriptInfo(BaseModel): + name: str = Field(default=None, title="Name", description="Script name") + is_alwayson: bool = Field(default=None, title="IsAlwayson", description="Flag specifying whether this script is an alwayson script") + is_img2img: bool = Field(default=None, title="IsImg2img", description="Flag specifying whether this script is an img2img script") + args: List[ScriptArg] = Field(title="Arguments", description="List of script's arguments") diff --git a/modules/call_queue.py b/modules/call_queue.py new file mode 100644 index 0000000000000000000000000000000000000000..ca56da5dfe1638dff78338ed5e488c82ed456f72 --- /dev/null +++ b/modules/call_queue.py @@ -0,0 +1,111 @@ +import html +import sys +import threading +import traceback +import time + +from modules import shared, progress + +queue_lock = threading.Lock() + + +def wrap_queued_call(func): + def f(*args, **kwargs): + with queue_lock: + res = func(*args, **kwargs) + + return res + + return f + + +def wrap_gradio_gpu_call(func, extra_outputs=None): + def f(*args, **kwargs): + + # if the first argument is a string that says "task(...)", it is treated as a job id + if len(args) > 0 and type(args[0]) == str and args[0][0:5] == "task(" and args[0][-1] == ")": + id_task = args[0] + progress.add_task_to_queue(id_task) + else: + id_task = None + + with queue_lock: + shared.state.begin() + progress.start_task(id_task) + + try: + res = func(*args, **kwargs) + progress.record_results(id_task, res) + finally: + progress.finish_task(id_task) + + shared.state.end() + + return res + + return wrap_gradio_call(f, extra_outputs=extra_outputs, add_stats=True) + + +def wrap_gradio_call(func, extra_outputs=None, add_stats=False): + def f(*args, extra_outputs_array=extra_outputs, **kwargs): + run_memmon = shared.opts.memmon_poll_rate > 0 and not shared.mem_mon.disabled and add_stats + if run_memmon: + shared.mem_mon.monitor() + t = time.perf_counter() + + try: + res = list(func(*args, **kwargs)) + except Exception as e: + # When printing out our debug argument list, do not print out more than a MB of text + max_debug_str_len = 131072 # (1024*1024)/8 + + print("Error completing request", file=sys.stderr) + argStr = f"Arguments: {args} {kwargs}" + print(argStr[:max_debug_str_len], file=sys.stderr) + if len(argStr) > max_debug_str_len: + print(f"(Argument list truncated at {max_debug_str_len}/{len(argStr)} characters)", file=sys.stderr) + + print(traceback.format_exc(), file=sys.stderr) + + shared.state.job = "" + shared.state.job_count = 0 + + if extra_outputs_array is None: + extra_outputs_array = [None, ''] + + error_message = f'{type(e).__name__}: {e}' + res = extra_outputs_array + [f"
{html.escape(error_message)}
"] + + shared.state.skipped = False + shared.state.interrupted = False + shared.state.job_count = 0 + + if not add_stats: + return tuple(res) + + elapsed = time.perf_counter() - t + elapsed_m = int(elapsed // 60) + elapsed_s = elapsed % 60 + elapsed_text = f"{elapsed_s:.2f}s" + if elapsed_m > 0: + elapsed_text = f"{elapsed_m}m "+elapsed_text + + if run_memmon: + mem_stats = {k: -(v//-(1024*1024)) for k, v in shared.mem_mon.stop().items()} + active_peak = mem_stats['active_peak'] + reserved_peak = mem_stats['reserved_peak'] + sys_peak = mem_stats['system_peak'] + sys_total = mem_stats['total'] + sys_pct = round(sys_peak/max(sys_total, 1) * 100, 2) + + vram_html = f"

Torch active/reserved: {active_peak}/{reserved_peak} MiB, Sys VRAM: {sys_peak}/{sys_total} MiB ({sys_pct}%)

" + else: + vram_html = '' + + # last item is always HTML + res[-1] += f"

Time taken: {elapsed_text}

{vram_html}
" + + return tuple(res) + + return f + diff --git a/modules/cmd_args.py b/modules/cmd_args.py new file mode 100644 index 0000000000000000000000000000000000000000..389bdd11ca791a45717e3326749004d5e4f4a1a0 --- /dev/null +++ b/modules/cmd_args.py @@ -0,0 +1,109 @@ +import argparse +import json +import os +from modules.paths_internal import models_path, script_path, data_path, extensions_dir, extensions_builtin_dir, sd_default_config, sd_model_file # noqa: F401 + +parser = argparse.ArgumentParser() + +parser.add_argument("-f", action='store_true', help=argparse.SUPPRESS) # allows running as root; implemented outside of webui +parser.add_argument("--update-all-extensions", action='store_true', help="launch.py argument: download updates for all extensions when starting the program") +parser.add_argument("--skip-python-version-check", action='store_true', help="launch.py argument: do not check python version") +parser.add_argument("--skip-torch-cuda-test", action='store_true', help="launch.py argument: do not check if CUDA is able to work properly") +parser.add_argument("--reinstall-xformers", action='store_true', help="launch.py argument: install the appropriate version of xformers even if you have some version already installed") +parser.add_argument("--reinstall-torch", action='store_true', help="launch.py argument: install the appropriate version of torch even if you have some version already installed") +parser.add_argument("--update-check", action='store_true', help="launch.py argument: chck for updates at startup") +parser.add_argument("--test-server", action='store_true', help="launch.py argument: configure server for testing") +parser.add_argument("--skip-prepare-environment", action='store_true', help="launch.py argument: skip all environment preparation") +parser.add_argument("--skip-install", action='store_true', help="launch.py argument: skip installation of packages") +parser.add_argument("--data-dir", type=str, default=os.path.dirname(os.path.dirname(os.path.realpath(__file__))), help="base path where all user data is stored") +parser.add_argument("--config", type=str, default=sd_default_config, help="path to config which constructs model",) +parser.add_argument("--ckpt", type=str, default=sd_model_file, help="path to checkpoint of stable diffusion model; if specified, this checkpoint will be added to the list of checkpoints and loaded",) +parser.add_argument("--ckpt-dir", type=str, default=None, help="Path to directory with stable diffusion checkpoints") +parser.add_argument("--vae-dir", type=str, default=None, help="Path to directory with VAE files") +parser.add_argument("--gfpgan-dir", type=str, help="GFPGAN directory", default=('./src/gfpgan' if os.path.exists('./src/gfpgan') else './GFPGAN')) +parser.add_argument("--gfpgan-model", type=str, help="GFPGAN model file name", default=None) +parser.add_argument("--no-half", action='store_true', help="do not switch the model to 16-bit floats") +parser.add_argument("--no-half-vae", action='store_true', help="do not switch the VAE model to 16-bit floats") +parser.add_argument("--no-progressbar-hiding", action='store_true', help="do not hide progressbar in gradio UI (we hide it because it slows down ML if you have hardware acceleration in browser)") +parser.add_argument("--max-batch-count", type=int, default=16, help="maximum batch count value for the UI") +parser.add_argument("--embeddings-dir", type=str, default=os.path.join(data_path, 'embeddings'), help="embeddings directory for textual inversion (default: embeddings)") +parser.add_argument("--textual-inversion-templates-dir", type=str, default=os.path.join(script_path, 'textual_inversion_templates'), help="directory with textual inversion templates") +parser.add_argument("--hypernetwork-dir", type=str, default=os.path.join(models_path, 'hypernetworks'), help="hypernetwork directory") +parser.add_argument("--localizations-dir", type=str, default=os.path.join(script_path, 'localizations'), help="localizations directory") +parser.add_argument("--allow-code", action='store_true', help="allow custom script execution from webui") +parser.add_argument("--medvram", action='store_true', help="enable stable diffusion model optimizations for sacrificing a little speed for low VRM usage") +parser.add_argument("--lowvram", action='store_true', help="enable stable diffusion model optimizations for sacrificing a lot of speed for very low VRM usage") +parser.add_argument("--lowram", action='store_true', help="load stable diffusion checkpoint weights to VRAM instead of RAM") +parser.add_argument("--always-batch-cond-uncond", action='store_true', help="disables cond/uncond batching that is enabled to save memory with --medvram or --lowvram") +parser.add_argument("--unload-gfpgan", action='store_true', help="does not do anything.") +parser.add_argument("--precision", type=str, help="evaluate at this precision", choices=["full", "autocast"], default="autocast") +parser.add_argument("--upcast-sampling", action='store_true', help="upcast sampling. No effect with --no-half. Usually produces similar results to --no-half with better performance while using less memory.") +parser.add_argument("--share", action='store_true', help="use share=True for gradio and make the UI accessible through their site") +parser.add_argument("--ngrok", type=str, help="ngrok authtoken, alternative to gradio --share", default=None) +parser.add_argument("--ngrok-region", type=str, help="does not do anything.", default="") +parser.add_argument("--ngrok-options", type=json.loads, help='The options to pass to ngrok in JSON format, e.g.: \'{"authtoken_from_env":true, "basic_auth":"user:password", "oauth_provider":"google", "oauth_allow_emails":"user@asdf.com"}\'', default=dict()) +parser.add_argument("--enable-insecure-extension-access", action='store_true', help="enable extensions tab regardless of other options") +parser.add_argument("--codeformer-models-path", type=str, help="Path to directory with codeformer model file(s).", default=os.path.join(models_path, 'Codeformer')) +parser.add_argument("--gfpgan-models-path", type=str, help="Path to directory with GFPGAN model file(s).", default=os.path.join(models_path, 'GFPGAN')) +parser.add_argument("--esrgan-models-path", type=str, help="Path to directory with ESRGAN model file(s).", default=os.path.join(models_path, 'ESRGAN')) +parser.add_argument("--bsrgan-models-path", type=str, help="Path to directory with BSRGAN model file(s).", default=os.path.join(models_path, 'BSRGAN')) +parser.add_argument("--realesrgan-models-path", type=str, help="Path to directory with RealESRGAN model file(s).", default=os.path.join(models_path, 'RealESRGAN')) +parser.add_argument("--clip-models-path", type=str, help="Path to directory with CLIP model file(s).", default=None) +parser.add_argument("--xformers", action='store_true', help="enable xformers for cross attention layers") +parser.add_argument("--force-enable-xformers", action='store_true', help="enable xformers for cross attention layers regardless of whether the checking code thinks you can run it; do not make bug reports if this fails to work") +parser.add_argument("--xformers-flash-attention", action='store_true', help="enable xformers with Flash Attention to improve reproducibility (supported for SD2.x or variant only)") +parser.add_argument("--deepdanbooru", action='store_true', help="does not do anything") +parser.add_argument("--opt-split-attention", action='store_true', help="prefer Doggettx's cross-attention layer optimization for automatic choice of optimization") +parser.add_argument("--opt-sub-quad-attention", action='store_true', help="prefer memory efficient sub-quadratic cross-attention layer optimization for automatic choice of optimization") +parser.add_argument("--sub-quad-q-chunk-size", type=int, help="query chunk size for the sub-quadratic cross-attention layer optimization to use", default=1024) +parser.add_argument("--sub-quad-kv-chunk-size", type=int, help="kv chunk size for the sub-quadratic cross-attention layer optimization to use", default=None) +parser.add_argument("--sub-quad-chunk-threshold", type=int, help="the percentage of VRAM threshold for the sub-quadratic cross-attention layer optimization to use chunking", default=None) +parser.add_argument("--opt-split-attention-invokeai", action='store_true', help="prefer InvokeAI's cross-attention layer optimization for automatic choice of optimization") +parser.add_argument("--opt-split-attention-v1", action='store_true', help="prefer older version of split attention optimization for automatic choice of optimization") +parser.add_argument("--opt-sdp-attention", action='store_true', help="prefer scaled dot product cross-attention layer optimization for automatic choice of optimization; requires PyTorch 2.*") +parser.add_argument("--opt-sdp-no-mem-attention", action='store_true', help="prefer scaled dot product cross-attention layer optimization without memory efficient attention for automatic choice of optimization, makes image generation deterministic; requires PyTorch 2.*") +parser.add_argument("--disable-opt-split-attention", action='store_true', help="prefer no cross-attention layer optimization for automatic choice of optimization") +parser.add_argument("--disable-nan-check", action='store_true', help="do not check if produced images/latent spaces have nans; useful for running without a checkpoint in CI") +parser.add_argument("--use-cpu", nargs='+', help="use CPU as torch device for specified modules", default=[], type=str.lower) +parser.add_argument("--listen", action='store_true', help="launch gradio with 0.0.0.0 as server name, allowing to respond to network requests") +parser.add_argument("--port", type=int, help="launch gradio with given server port, you need root/admin rights for ports < 1024, defaults to 7860 if available", default=None) +parser.add_argument("--show-negative-prompt", action='store_true', help="does not do anything", default=False) +parser.add_argument("--ui-config-file", type=str, help="filename to use for ui configuration", default=os.path.join(data_path, 'ui-config.json')) +parser.add_argument("--hide-ui-dir-config", action='store_true', help="hide directory configuration from webui", default=False) +parser.add_argument("--freeze-settings", action='store_true', help="disable editing settings", default=False) +parser.add_argument("--ui-settings-file", type=str, help="filename to use for ui settings", default=os.path.join(data_path, 'config.json')) +parser.add_argument("--gradio-debug", action='store_true', help="launch gradio with --debug option") +parser.add_argument("--gradio-auth", type=str, help='set gradio authentication like "username:password"; or comma-delimit multiple like "u1:p1,u2:p2,u3:p3"', default=None) +parser.add_argument("--gradio-auth-path", type=str, help='set gradio authentication file path ex. "/path/to/auth/file" same auth format as --gradio-auth', default=None) +parser.add_argument("--gradio-img2img-tool", type=str, help='does not do anything') +parser.add_argument("--gradio-inpaint-tool", type=str, help="does not do anything") +parser.add_argument("--gradio-allowed-path", action='append', help="add path to gradio's allowed_paths, make it possible to serve files from it") +parser.add_argument("--opt-channelslast", action='store_true', help="change memory type for stable diffusion to channels last") +parser.add_argument("--styles-file", type=str, help="filename to use for styles", default=os.path.join(data_path, 'styles.csv')) +parser.add_argument("--autolaunch", action='store_true', help="open the webui URL in the system's default browser upon launch", default=False) +parser.add_argument("--theme", type=str, help="launches the UI with light or dark theme", default=None) +parser.add_argument("--use-textbox-seed", action='store_true', help="use textbox for seeds in UI (no up/down, but possible to input long seeds)", default=False) +parser.add_argument("--disable-console-progressbars", action='store_true', help="do not output progressbars to console", default=False) +parser.add_argument("--enable-console-prompts", action='store_true', help="print prompts to console when generating with txt2img and img2img", default=False) +parser.add_argument('--vae-path', type=str, help='Checkpoint to use as VAE; setting this argument disables all settings related to VAE', default=None) +parser.add_argument("--disable-safe-unpickle", action='store_true', help="disable checking pytorch models for malicious code", default=False) +parser.add_argument("--api", action='store_true', help="use api=True to launch the API together with the webui (use --nowebui instead for only the API)") +parser.add_argument("--api-auth", type=str, help='Set authentication for API like "username:password"; or comma-delimit multiple like "u1:p1,u2:p2,u3:p3"', default=None) +parser.add_argument("--api-log", action='store_true', help="use api-log=True to enable logging of all API requests") +parser.add_argument("--nowebui", action='store_true', help="use api=True to launch the API instead of the webui") +parser.add_argument("--ui-debug-mode", action='store_true', help="Don't load model to quickly launch UI") +parser.add_argument("--device-id", type=str, help="Select the default CUDA device to use (export CUDA_VISIBLE_DEVICES=0,1,etc might be needed before)", default=None) +parser.add_argument("--administrator", action='store_true', help="Administrator rights", default=False) +parser.add_argument("--cors-allow-origins", type=str, help="Allowed CORS origin(s) in the form of a comma-separated list (no spaces)", default=None) +parser.add_argument("--cors-allow-origins-regex", type=str, help="Allowed CORS origin(s) in the form of a single regular expression", default=None) +parser.add_argument("--tls-keyfile", type=str, help="Partially enables TLS, requires --tls-certfile to fully function", default=None) +parser.add_argument("--tls-certfile", type=str, help="Partially enables TLS, requires --tls-keyfile to fully function", default=None) +parser.add_argument("--disable-tls-verify", action="store_false", help="When passed, enables the use of self-signed certificates.", default=None) +parser.add_argument("--server-name", type=str, help="Sets hostname of server", default=None) +parser.add_argument("--gradio-queue", action='store_true', help="does not do anything", default=True) +parser.add_argument("--no-gradio-queue", action='store_true', help="Disables gradio queue; causes the webpage to use http requests instead of websockets; was the defaul in earlier versions") +parser.add_argument("--skip-version-check", action='store_true', help="Do not check versions of torch and xformers") +parser.add_argument("--no-hashing", action='store_true', help="disable sha256 hashing of checkpoints to help loading performance", default=False) +parser.add_argument("--no-download-sd-model", action='store_true', help="don't download SD1.5 model even if no model is found in --ckpt-dir", default=False) +parser.add_argument('--subpath', type=str, help='customize the subpath for gradio, use with reverse proxy') +parser.add_argument('--add-stop-route', action='store_true', help='add /_stop route to stop server') diff --git a/modules/codeformer/codeformer_arch.py b/modules/codeformer/codeformer_arch.py new file mode 100644 index 0000000000000000000000000000000000000000..12db6814268fdba5a3025f44d1bb24e93d280a69 --- /dev/null +++ b/modules/codeformer/codeformer_arch.py @@ -0,0 +1,276 @@ +# this file is copied from CodeFormer repository. Please see comment in modules/codeformer_model.py + +import math +import torch +from torch import nn, Tensor +import torch.nn.functional as F +from typing import Optional + +from modules.codeformer.vqgan_arch import VQAutoEncoder, ResBlock +from basicsr.utils.registry import ARCH_REGISTRY + +def calc_mean_std(feat, eps=1e-5): + """Calculate mean and std for adaptive_instance_normalization. + + Args: + feat (Tensor): 4D tensor. + eps (float): A small value added to the variance to avoid + divide-by-zero. Default: 1e-5. + """ + size = feat.size() + assert len(size) == 4, 'The input feature should be 4D tensor.' + b, c = size[:2] + feat_var = feat.view(b, c, -1).var(dim=2) + eps + feat_std = feat_var.sqrt().view(b, c, 1, 1) + feat_mean = feat.view(b, c, -1).mean(dim=2).view(b, c, 1, 1) + return feat_mean, feat_std + + +def adaptive_instance_normalization(content_feat, style_feat): + """Adaptive instance normalization. + + Adjust the reference features to have the similar color and illuminations + as those in the degradate features. + + Args: + content_feat (Tensor): The reference feature. + style_feat (Tensor): The degradate features. + """ + size = content_feat.size() + style_mean, style_std = calc_mean_std(style_feat) + content_mean, content_std = calc_mean_std(content_feat) + normalized_feat = (content_feat - content_mean.expand(size)) / content_std.expand(size) + return normalized_feat * style_std.expand(size) + style_mean.expand(size) + + +class PositionEmbeddingSine(nn.Module): + """ + This is a more standard version of the position embedding, very similar to the one + used by the Attention is all you need paper, generalized to work on images. + """ + + def __init__(self, num_pos_feats=64, temperature=10000, normalize=False, scale=None): + super().__init__() + self.num_pos_feats = num_pos_feats + self.temperature = temperature + self.normalize = normalize + if scale is not None and normalize is False: + raise ValueError("normalize should be True if scale is passed") + if scale is None: + scale = 2 * math.pi + self.scale = scale + + def forward(self, x, mask=None): + if mask is None: + mask = torch.zeros((x.size(0), x.size(2), x.size(3)), device=x.device, dtype=torch.bool) + not_mask = ~mask + y_embed = not_mask.cumsum(1, dtype=torch.float32) + x_embed = not_mask.cumsum(2, dtype=torch.float32) + if self.normalize: + eps = 1e-6 + y_embed = y_embed / (y_embed[:, -1:, :] + eps) * self.scale + x_embed = x_embed / (x_embed[:, :, -1:] + eps) * self.scale + + dim_t = torch.arange(self.num_pos_feats, dtype=torch.float32, device=x.device) + dim_t = self.temperature ** (2 * (dim_t // 2) / self.num_pos_feats) + + pos_x = x_embed[:, :, :, None] / dim_t + pos_y = y_embed[:, :, :, None] / dim_t + pos_x = torch.stack( + (pos_x[:, :, :, 0::2].sin(), pos_x[:, :, :, 1::2].cos()), dim=4 + ).flatten(3) + pos_y = torch.stack( + (pos_y[:, :, :, 0::2].sin(), pos_y[:, :, :, 1::2].cos()), dim=4 + ).flatten(3) + pos = torch.cat((pos_y, pos_x), dim=3).permute(0, 3, 1, 2) + return pos + +def _get_activation_fn(activation): + """Return an activation function given a string""" + if activation == "relu": + return F.relu + if activation == "gelu": + return F.gelu + if activation == "glu": + return F.glu + raise RuntimeError(F"activation should be relu/gelu, not {activation}.") + + +class TransformerSALayer(nn.Module): + def __init__(self, embed_dim, nhead=8, dim_mlp=2048, dropout=0.0, activation="gelu"): + super().__init__() + self.self_attn = nn.MultiheadAttention(embed_dim, nhead, dropout=dropout) + # Implementation of Feedforward model - MLP + self.linear1 = nn.Linear(embed_dim, dim_mlp) + self.dropout = nn.Dropout(dropout) + self.linear2 = nn.Linear(dim_mlp, embed_dim) + + self.norm1 = nn.LayerNorm(embed_dim) + self.norm2 = nn.LayerNorm(embed_dim) + self.dropout1 = nn.Dropout(dropout) + self.dropout2 = nn.Dropout(dropout) + + self.activation = _get_activation_fn(activation) + + def with_pos_embed(self, tensor, pos: Optional[Tensor]): + return tensor if pos is None else tensor + pos + + def forward(self, tgt, + tgt_mask: Optional[Tensor] = None, + tgt_key_padding_mask: Optional[Tensor] = None, + query_pos: Optional[Tensor] = None): + + # self attention + tgt2 = self.norm1(tgt) + q = k = self.with_pos_embed(tgt2, query_pos) + tgt2 = self.self_attn(q, k, value=tgt2, attn_mask=tgt_mask, + key_padding_mask=tgt_key_padding_mask)[0] + tgt = tgt + self.dropout1(tgt2) + + # ffn + tgt2 = self.norm2(tgt) + tgt2 = self.linear2(self.dropout(self.activation(self.linear1(tgt2)))) + tgt = tgt + self.dropout2(tgt2) + return tgt + +class Fuse_sft_block(nn.Module): + def __init__(self, in_ch, out_ch): + super().__init__() + self.encode_enc = ResBlock(2*in_ch, out_ch) + + self.scale = nn.Sequential( + nn.Conv2d(in_ch, out_ch, kernel_size=3, padding=1), + nn.LeakyReLU(0.2, True), + nn.Conv2d(out_ch, out_ch, kernel_size=3, padding=1)) + + self.shift = nn.Sequential( + nn.Conv2d(in_ch, out_ch, kernel_size=3, padding=1), + nn.LeakyReLU(0.2, True), + nn.Conv2d(out_ch, out_ch, kernel_size=3, padding=1)) + + def forward(self, enc_feat, dec_feat, w=1): + enc_feat = self.encode_enc(torch.cat([enc_feat, dec_feat], dim=1)) + scale = self.scale(enc_feat) + shift = self.shift(enc_feat) + residual = w * (dec_feat * scale + shift) + out = dec_feat + residual + return out + + +@ARCH_REGISTRY.register() +class CodeFormer(VQAutoEncoder): + def __init__(self, dim_embd=512, n_head=8, n_layers=9, + codebook_size=1024, latent_size=256, + connect_list=('32', '64', '128', '256'), + fix_modules=('quantize', 'generator')): + super(CodeFormer, self).__init__(512, 64, [1, 2, 2, 4, 4, 8], 'nearest',2, [16], codebook_size) + + if fix_modules is not None: + for module in fix_modules: + for param in getattr(self, module).parameters(): + param.requires_grad = False + + self.connect_list = connect_list + self.n_layers = n_layers + self.dim_embd = dim_embd + self.dim_mlp = dim_embd*2 + + self.position_emb = nn.Parameter(torch.zeros(latent_size, self.dim_embd)) + self.feat_emb = nn.Linear(256, self.dim_embd) + + # transformer + self.ft_layers = nn.Sequential(*[TransformerSALayer(embed_dim=dim_embd, nhead=n_head, dim_mlp=self.dim_mlp, dropout=0.0) + for _ in range(self.n_layers)]) + + # logits_predict head + self.idx_pred_layer = nn.Sequential( + nn.LayerNorm(dim_embd), + nn.Linear(dim_embd, codebook_size, bias=False)) + + self.channels = { + '16': 512, + '32': 256, + '64': 256, + '128': 128, + '256': 128, + '512': 64, + } + + # after second residual block for > 16, before attn layer for ==16 + self.fuse_encoder_block = {'512':2, '256':5, '128':8, '64':11, '32':14, '16':18} + # after first residual block for > 16, before attn layer for ==16 + self.fuse_generator_block = {'16':6, '32': 9, '64':12, '128':15, '256':18, '512':21} + + # fuse_convs_dict + self.fuse_convs_dict = nn.ModuleDict() + for f_size in self.connect_list: + in_ch = self.channels[f_size] + self.fuse_convs_dict[f_size] = Fuse_sft_block(in_ch, in_ch) + + def _init_weights(self, module): + if isinstance(module, (nn.Linear, nn.Embedding)): + module.weight.data.normal_(mean=0.0, std=0.02) + if isinstance(module, nn.Linear) and module.bias is not None: + module.bias.data.zero_() + elif isinstance(module, nn.LayerNorm): + module.bias.data.zero_() + module.weight.data.fill_(1.0) + + def forward(self, x, w=0, detach_16=True, code_only=False, adain=False): + # ################### Encoder ##################### + enc_feat_dict = {} + out_list = [self.fuse_encoder_block[f_size] for f_size in self.connect_list] + for i, block in enumerate(self.encoder.blocks): + x = block(x) + if i in out_list: + enc_feat_dict[str(x.shape[-1])] = x.clone() + + lq_feat = x + # ################# Transformer ################### + # quant_feat, codebook_loss, quant_stats = self.quantize(lq_feat) + pos_emb = self.position_emb.unsqueeze(1).repeat(1,x.shape[0],1) + # BCHW -> BC(HW) -> (HW)BC + feat_emb = self.feat_emb(lq_feat.flatten(2).permute(2,0,1)) + query_emb = feat_emb + # Transformer encoder + for layer in self.ft_layers: + query_emb = layer(query_emb, query_pos=pos_emb) + + # output logits + logits = self.idx_pred_layer(query_emb) # (hw)bn + logits = logits.permute(1,0,2) # (hw)bn -> b(hw)n + + if code_only: # for training stage II + # logits doesn't need softmax before cross_entropy loss + return logits, lq_feat + + # ################# Quantization ################### + # if self.training: + # quant_feat = torch.einsum('btn,nc->btc', [soft_one_hot, self.quantize.embedding.weight]) + # # b(hw)c -> bc(hw) -> bchw + # quant_feat = quant_feat.permute(0,2,1).view(lq_feat.shape) + # ------------ + soft_one_hot = F.softmax(logits, dim=2) + _, top_idx = torch.topk(soft_one_hot, 1, dim=2) + quant_feat = self.quantize.get_codebook_feat(top_idx, shape=[x.shape[0],16,16,256]) + # preserve gradients + # quant_feat = lq_feat + (quant_feat - lq_feat).detach() + + if detach_16: + quant_feat = quant_feat.detach() # for training stage III + if adain: + quant_feat = adaptive_instance_normalization(quant_feat, lq_feat) + + # ################## Generator #################### + x = quant_feat + fuse_list = [self.fuse_generator_block[f_size] for f_size in self.connect_list] + + for i, block in enumerate(self.generator.blocks): + x = block(x) + if i in fuse_list: # fuse after i-th block + f_size = str(x.shape[-1]) + if w>0: + x = self.fuse_convs_dict[f_size](enc_feat_dict[f_size].detach(), x, w) + out = x + # logits doesn't need softmax before cross_entropy loss + return out, logits, lq_feat diff --git a/modules/codeformer/vqgan_arch.py b/modules/codeformer/vqgan_arch.py new file mode 100644 index 0000000000000000000000000000000000000000..09ee6660dc537e41fb9d9c7be7196c94c04aa8c6 --- /dev/null +++ b/modules/codeformer/vqgan_arch.py @@ -0,0 +1,435 @@ +# this file is copied from CodeFormer repository. Please see comment in modules/codeformer_model.py + +''' +VQGAN code, adapted from the original created by the Unleashing Transformers authors: +https://github.com/samb-t/unleashing-transformers/blob/master/models/vqgan.py + +''' +import torch +import torch.nn as nn +import torch.nn.functional as F +from basicsr.utils import get_root_logger +from basicsr.utils.registry import ARCH_REGISTRY + +def normalize(in_channels): + return torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True) + + +@torch.jit.script +def swish(x): + return x*torch.sigmoid(x) + + +# Define VQVAE classes +class VectorQuantizer(nn.Module): + def __init__(self, codebook_size, emb_dim, beta): + super(VectorQuantizer, self).__init__() + self.codebook_size = codebook_size # number of embeddings + self.emb_dim = emb_dim # dimension of embedding + self.beta = beta # commitment cost used in loss term, beta * ||z_e(x)-sg[e]||^2 + self.embedding = nn.Embedding(self.codebook_size, self.emb_dim) + self.embedding.weight.data.uniform_(-1.0 / self.codebook_size, 1.0 / self.codebook_size) + + def forward(self, z): + # reshape z -> (batch, height, width, channel) and flatten + z = z.permute(0, 2, 3, 1).contiguous() + z_flattened = z.view(-1, self.emb_dim) + + # distances from z to embeddings e_j (z - e)^2 = z^2 + e^2 - 2 e * z + d = (z_flattened ** 2).sum(dim=1, keepdim=True) + (self.embedding.weight**2).sum(1) - \ + 2 * torch.matmul(z_flattened, self.embedding.weight.t()) + + mean_distance = torch.mean(d) + # find closest encodings + # min_encoding_indices = torch.argmin(d, dim=1).unsqueeze(1) + min_encoding_scores, min_encoding_indices = torch.topk(d, 1, dim=1, largest=False) + # [0-1], higher score, higher confidence + min_encoding_scores = torch.exp(-min_encoding_scores/10) + + min_encodings = torch.zeros(min_encoding_indices.shape[0], self.codebook_size).to(z) + min_encodings.scatter_(1, min_encoding_indices, 1) + + # get quantized latent vectors + z_q = torch.matmul(min_encodings, self.embedding.weight).view(z.shape) + # compute loss for embedding + loss = torch.mean((z_q.detach()-z)**2) + self.beta * torch.mean((z_q - z.detach()) ** 2) + # preserve gradients + z_q = z + (z_q - z).detach() + + # perplexity + e_mean = torch.mean(min_encodings, dim=0) + perplexity = torch.exp(-torch.sum(e_mean * torch.log(e_mean + 1e-10))) + # reshape back to match original input shape + z_q = z_q.permute(0, 3, 1, 2).contiguous() + + return z_q, loss, { + "perplexity": perplexity, + "min_encodings": min_encodings, + "min_encoding_indices": min_encoding_indices, + "min_encoding_scores": min_encoding_scores, + "mean_distance": mean_distance + } + + def get_codebook_feat(self, indices, shape): + # input indices: batch*token_num -> (batch*token_num)*1 + # shape: batch, height, width, channel + indices = indices.view(-1,1) + min_encodings = torch.zeros(indices.shape[0], self.codebook_size).to(indices) + min_encodings.scatter_(1, indices, 1) + # get quantized latent vectors + z_q = torch.matmul(min_encodings.float(), self.embedding.weight) + + if shape is not None: # reshape back to match original input shape + z_q = z_q.view(shape).permute(0, 3, 1, 2).contiguous() + + return z_q + + +class GumbelQuantizer(nn.Module): + def __init__(self, codebook_size, emb_dim, num_hiddens, straight_through=False, kl_weight=5e-4, temp_init=1.0): + super().__init__() + self.codebook_size = codebook_size # number of embeddings + self.emb_dim = emb_dim # dimension of embedding + self.straight_through = straight_through + self.temperature = temp_init + self.kl_weight = kl_weight + self.proj = nn.Conv2d(num_hiddens, codebook_size, 1) # projects last encoder layer to quantized logits + self.embed = nn.Embedding(codebook_size, emb_dim) + + def forward(self, z): + hard = self.straight_through if self.training else True + + logits = self.proj(z) + + soft_one_hot = F.gumbel_softmax(logits, tau=self.temperature, dim=1, hard=hard) + + z_q = torch.einsum("b n h w, n d -> b d h w", soft_one_hot, self.embed.weight) + + # + kl divergence to the prior loss + qy = F.softmax(logits, dim=1) + diff = self.kl_weight * torch.sum(qy * torch.log(qy * self.codebook_size + 1e-10), dim=1).mean() + min_encoding_indices = soft_one_hot.argmax(dim=1) + + return z_q, diff, { + "min_encoding_indices": min_encoding_indices + } + + +class Downsample(nn.Module): + def __init__(self, in_channels): + super().__init__() + self.conv = torch.nn.Conv2d(in_channels, in_channels, kernel_size=3, stride=2, padding=0) + + def forward(self, x): + pad = (0, 1, 0, 1) + x = torch.nn.functional.pad(x, pad, mode="constant", value=0) + x = self.conv(x) + return x + + +class Upsample(nn.Module): + def __init__(self, in_channels): + super().__init__() + self.conv = nn.Conv2d(in_channels, in_channels, kernel_size=3, stride=1, padding=1) + + def forward(self, x): + x = F.interpolate(x, scale_factor=2.0, mode="nearest") + x = self.conv(x) + + return x + + +class ResBlock(nn.Module): + def __init__(self, in_channels, out_channels=None): + super(ResBlock, self).__init__() + self.in_channels = in_channels + self.out_channels = in_channels if out_channels is None else out_channels + self.norm1 = normalize(in_channels) + self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, stride=1, padding=1) + self.norm2 = normalize(out_channels) + self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, stride=1, padding=1) + if self.in_channels != self.out_channels: + self.conv_out = nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=1, padding=0) + + def forward(self, x_in): + x = x_in + x = self.norm1(x) + x = swish(x) + x = self.conv1(x) + x = self.norm2(x) + x = swish(x) + x = self.conv2(x) + if self.in_channels != self.out_channels: + x_in = self.conv_out(x_in) + + return x + x_in + + +class AttnBlock(nn.Module): + def __init__(self, in_channels): + super().__init__() + self.in_channels = in_channels + + self.norm = normalize(in_channels) + self.q = torch.nn.Conv2d( + in_channels, + in_channels, + kernel_size=1, + stride=1, + padding=0 + ) + self.k = torch.nn.Conv2d( + in_channels, + in_channels, + kernel_size=1, + stride=1, + padding=0 + ) + self.v = torch.nn.Conv2d( + in_channels, + in_channels, + kernel_size=1, + stride=1, + padding=0 + ) + self.proj_out = torch.nn.Conv2d( + in_channels, + in_channels, + kernel_size=1, + stride=1, + padding=0 + ) + + def forward(self, x): + h_ = x + h_ = self.norm(h_) + q = self.q(h_) + k = self.k(h_) + v = self.v(h_) + + # compute attention + b, c, h, w = q.shape + q = q.reshape(b, c, h*w) + q = q.permute(0, 2, 1) + k = k.reshape(b, c, h*w) + w_ = torch.bmm(q, k) + w_ = w_ * (int(c)**(-0.5)) + w_ = F.softmax(w_, dim=2) + + # attend to values + v = v.reshape(b, c, h*w) + w_ = w_.permute(0, 2, 1) + h_ = torch.bmm(v, w_) + h_ = h_.reshape(b, c, h, w) + + h_ = self.proj_out(h_) + + return x+h_ + + +class Encoder(nn.Module): + def __init__(self, in_channels, nf, emb_dim, ch_mult, num_res_blocks, resolution, attn_resolutions): + super().__init__() + self.nf = nf + self.num_resolutions = len(ch_mult) + self.num_res_blocks = num_res_blocks + self.resolution = resolution + self.attn_resolutions = attn_resolutions + + curr_res = self.resolution + in_ch_mult = (1,)+tuple(ch_mult) + + blocks = [] + # initial convultion + blocks.append(nn.Conv2d(in_channels, nf, kernel_size=3, stride=1, padding=1)) + + # residual and downsampling blocks, with attention on smaller res (16x16) + for i in range(self.num_resolutions): + block_in_ch = nf * in_ch_mult[i] + block_out_ch = nf * ch_mult[i] + for _ in range(self.num_res_blocks): + blocks.append(ResBlock(block_in_ch, block_out_ch)) + block_in_ch = block_out_ch + if curr_res in attn_resolutions: + blocks.append(AttnBlock(block_in_ch)) + + if i != self.num_resolutions - 1: + blocks.append(Downsample(block_in_ch)) + curr_res = curr_res // 2 + + # non-local attention block + blocks.append(ResBlock(block_in_ch, block_in_ch)) + blocks.append(AttnBlock(block_in_ch)) + blocks.append(ResBlock(block_in_ch, block_in_ch)) + + # normalise and convert to latent size + blocks.append(normalize(block_in_ch)) + blocks.append(nn.Conv2d(block_in_ch, emb_dim, kernel_size=3, stride=1, padding=1)) + self.blocks = nn.ModuleList(blocks) + + def forward(self, x): + for block in self.blocks: + x = block(x) + + return x + + +class Generator(nn.Module): + def __init__(self, nf, emb_dim, ch_mult, res_blocks, img_size, attn_resolutions): + super().__init__() + self.nf = nf + self.ch_mult = ch_mult + self.num_resolutions = len(self.ch_mult) + self.num_res_blocks = res_blocks + self.resolution = img_size + self.attn_resolutions = attn_resolutions + self.in_channels = emb_dim + self.out_channels = 3 + block_in_ch = self.nf * self.ch_mult[-1] + curr_res = self.resolution // 2 ** (self.num_resolutions-1) + + blocks = [] + # initial conv + blocks.append(nn.Conv2d(self.in_channels, block_in_ch, kernel_size=3, stride=1, padding=1)) + + # non-local attention block + blocks.append(ResBlock(block_in_ch, block_in_ch)) + blocks.append(AttnBlock(block_in_ch)) + blocks.append(ResBlock(block_in_ch, block_in_ch)) + + for i in reversed(range(self.num_resolutions)): + block_out_ch = self.nf * self.ch_mult[i] + + for _ in range(self.num_res_blocks): + blocks.append(ResBlock(block_in_ch, block_out_ch)) + block_in_ch = block_out_ch + + if curr_res in self.attn_resolutions: + blocks.append(AttnBlock(block_in_ch)) + + if i != 0: + blocks.append(Upsample(block_in_ch)) + curr_res = curr_res * 2 + + blocks.append(normalize(block_in_ch)) + blocks.append(nn.Conv2d(block_in_ch, self.out_channels, kernel_size=3, stride=1, padding=1)) + + self.blocks = nn.ModuleList(blocks) + + + def forward(self, x): + for block in self.blocks: + x = block(x) + + return x + + +@ARCH_REGISTRY.register() +class VQAutoEncoder(nn.Module): + def __init__(self, img_size, nf, ch_mult, quantizer="nearest", res_blocks=2, attn_resolutions=None, codebook_size=1024, emb_dim=256, + beta=0.25, gumbel_straight_through=False, gumbel_kl_weight=1e-8, model_path=None): + super().__init__() + logger = get_root_logger() + self.in_channels = 3 + self.nf = nf + self.n_blocks = res_blocks + self.codebook_size = codebook_size + self.embed_dim = emb_dim + self.ch_mult = ch_mult + self.resolution = img_size + self.attn_resolutions = attn_resolutions or [16] + self.quantizer_type = quantizer + self.encoder = Encoder( + self.in_channels, + self.nf, + self.embed_dim, + self.ch_mult, + self.n_blocks, + self.resolution, + self.attn_resolutions + ) + if self.quantizer_type == "nearest": + self.beta = beta #0.25 + self.quantize = VectorQuantizer(self.codebook_size, self.embed_dim, self.beta) + elif self.quantizer_type == "gumbel": + self.gumbel_num_hiddens = emb_dim + self.straight_through = gumbel_straight_through + self.kl_weight = gumbel_kl_weight + self.quantize = GumbelQuantizer( + self.codebook_size, + self.embed_dim, + self.gumbel_num_hiddens, + self.straight_through, + self.kl_weight + ) + self.generator = Generator( + self.nf, + self.embed_dim, + self.ch_mult, + self.n_blocks, + self.resolution, + self.attn_resolutions + ) + + if model_path is not None: + chkpt = torch.load(model_path, map_location='cpu') + if 'params_ema' in chkpt: + self.load_state_dict(torch.load(model_path, map_location='cpu')['params_ema']) + logger.info(f'vqgan is loaded from: {model_path} [params_ema]') + elif 'params' in chkpt: + self.load_state_dict(torch.load(model_path, map_location='cpu')['params']) + logger.info(f'vqgan is loaded from: {model_path} [params]') + else: + raise ValueError('Wrong params!') + + + def forward(self, x): + x = self.encoder(x) + quant, codebook_loss, quant_stats = self.quantize(x) + x = self.generator(quant) + return x, codebook_loss, quant_stats + + + +# patch based discriminator +@ARCH_REGISTRY.register() +class VQGANDiscriminator(nn.Module): + def __init__(self, nc=3, ndf=64, n_layers=4, model_path=None): + super().__init__() + + layers = [nn.Conv2d(nc, ndf, kernel_size=4, stride=2, padding=1), nn.LeakyReLU(0.2, True)] + ndf_mult = 1 + ndf_mult_prev = 1 + for n in range(1, n_layers): # gradually increase the number of filters + ndf_mult_prev = ndf_mult + ndf_mult = min(2 ** n, 8) + layers += [ + nn.Conv2d(ndf * ndf_mult_prev, ndf * ndf_mult, kernel_size=4, stride=2, padding=1, bias=False), + nn.BatchNorm2d(ndf * ndf_mult), + nn.LeakyReLU(0.2, True) + ] + + ndf_mult_prev = ndf_mult + ndf_mult = min(2 ** n_layers, 8) + + layers += [ + nn.Conv2d(ndf * ndf_mult_prev, ndf * ndf_mult, kernel_size=4, stride=1, padding=1, bias=False), + nn.BatchNorm2d(ndf * ndf_mult), + nn.LeakyReLU(0.2, True) + ] + + layers += [ + nn.Conv2d(ndf * ndf_mult, 1, kernel_size=4, stride=1, padding=1)] # output 1 channel prediction map + self.main = nn.Sequential(*layers) + + if model_path is not None: + chkpt = torch.load(model_path, map_location='cpu') + if 'params_d' in chkpt: + self.load_state_dict(torch.load(model_path, map_location='cpu')['params_d']) + elif 'params' in chkpt: + self.load_state_dict(torch.load(model_path, map_location='cpu')['params']) + else: + raise ValueError('Wrong params!') + + def forward(self, x): + return self.main(x) diff --git a/modules/codeformer_model.py b/modules/codeformer_model.py new file mode 100644 index 0000000000000000000000000000000000000000..9bb3ae83ee9e61cbdf695a3f43034c116c68a01f --- /dev/null +++ b/modules/codeformer_model.py @@ -0,0 +1,141 @@ +import os +import sys +import traceback + +import cv2 +import torch + +import modules.face_restoration +import modules.shared +from modules import shared, devices, modelloader +from modules.paths import models_path + +# codeformer people made a choice to include modified basicsr library to their project which makes +# it utterly impossible to use it alongside with other libraries that also use basicsr, like GFPGAN. +# I am making a choice to include some files from codeformer to work around this issue. +model_dir = "Codeformer" +model_path = os.path.join(models_path, model_dir) +model_url = 'https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/codeformer.pth' + +have_codeformer = False +codeformer = None + + +def setup_model(dirname): + global model_path + if not os.path.exists(model_path): + os.makedirs(model_path) + + path = modules.paths.paths.get("CodeFormer", None) + if path is None: + return + + try: + from torchvision.transforms.functional import normalize + from modules.codeformer.codeformer_arch import CodeFormer + from basicsr.utils import img2tensor, tensor2img + from facelib.utils.face_restoration_helper import FaceRestoreHelper + from facelib.detection.retinaface import retinaface + + net_class = CodeFormer + + class FaceRestorerCodeFormer(modules.face_restoration.FaceRestoration): + def name(self): + return "CodeFormer" + + def __init__(self, dirname): + self.net = None + self.face_helper = None + self.cmd_dir = dirname + + def create_models(self): + + if self.net is not None and self.face_helper is not None: + self.net.to(devices.device_codeformer) + return self.net, self.face_helper + model_paths = modelloader.load_models(model_path, model_url, self.cmd_dir, download_name='codeformer-v0.1.0.pth', ext_filter=['.pth']) + if len(model_paths) != 0: + ckpt_path = model_paths[0] + else: + print("Unable to load codeformer model.") + return None, None + net = net_class(dim_embd=512, codebook_size=1024, n_head=8, n_layers=9, connect_list=['32', '64', '128', '256']).to(devices.device_codeformer) + checkpoint = torch.load(ckpt_path)['params_ema'] + net.load_state_dict(checkpoint) + net.eval() + + if hasattr(retinaface, 'device'): + retinaface.device = devices.device_codeformer + face_helper = FaceRestoreHelper(1, face_size=512, crop_ratio=(1, 1), det_model='retinaface_resnet50', save_ext='png', use_parse=True, device=devices.device_codeformer) + + self.net = net + self.face_helper = face_helper + + return net, face_helper + + def send_model_to(self, device): + self.net.to(device) + self.face_helper.face_det.to(device) + self.face_helper.face_parse.to(device) + + def restore(self, np_image, w=None): + np_image = np_image[:, :, ::-1] + + original_resolution = np_image.shape[0:2] + + self.create_models() + if self.net is None or self.face_helper is None: + return np_image + + self.send_model_to(devices.device_codeformer) + + self.face_helper.clean_all() + self.face_helper.read_image(np_image) + self.face_helper.get_face_landmarks_5(only_center_face=False, resize=640, eye_dist_threshold=5) + self.face_helper.align_warp_face() + + for cropped_face in self.face_helper.cropped_faces: + cropped_face_t = img2tensor(cropped_face / 255., bgr2rgb=True, float32=True) + normalize(cropped_face_t, (0.5, 0.5, 0.5), (0.5, 0.5, 0.5), inplace=True) + cropped_face_t = cropped_face_t.unsqueeze(0).to(devices.device_codeformer) + + try: + with torch.no_grad(): + output = self.net(cropped_face_t, w=w if w is not None else shared.opts.code_former_weight, adain=True)[0] + restored_face = tensor2img(output, rgb2bgr=True, min_max=(-1, 1)) + del output + torch.cuda.empty_cache() + except Exception as error: + print(f'\tFailed inference for CodeFormer: {error}', file=sys.stderr) + restored_face = tensor2img(cropped_face_t, rgb2bgr=True, min_max=(-1, 1)) + + restored_face = restored_face.astype('uint8') + self.face_helper.add_restored_face(restored_face) + + self.face_helper.get_inverse_affine(None) + + restored_img = self.face_helper.paste_faces_to_input_image() + restored_img = restored_img[:, :, ::-1] + + if original_resolution != restored_img.shape[0:2]: + restored_img = cv2.resize(restored_img, (0, 0), fx=original_resolution[1]/restored_img.shape[1], fy=original_resolution[0]/restored_img.shape[0], interpolation=cv2.INTER_LINEAR) + + self.face_helper.clean_all() + + if shared.opts.face_restoration_unload: + self.send_model_to(devices.cpu) + + return restored_img + + global have_codeformer + have_codeformer = True + + global codeformer + codeformer = FaceRestorerCodeFormer(dirname) + shared.face_restorers.append(codeformer) + + except Exception: + print("Error setting up CodeFormer:", file=sys.stderr) + print(traceback.format_exc(), file=sys.stderr) + + # sys.path = stored_sys_path diff --git a/modules/config_states.py b/modules/config_states.py new file mode 100644 index 0000000000000000000000000000000000000000..db65bcdbf7a8f484dddde20ba7b54ef9b8e0ecd4 --- /dev/null +++ b/modules/config_states.py @@ -0,0 +1,202 @@ +""" +Supports saving and restoring webui and extensions from a known working set of commits +""" + +import os +import sys +import traceback +import json +import time +import tqdm + +from datetime import datetime +from collections import OrderedDict +import git + +from modules import shared, extensions +from modules.paths_internal import script_path, config_states_dir + + +all_config_states = OrderedDict() + + +def list_config_states(): + global all_config_states + + all_config_states.clear() + os.makedirs(config_states_dir, exist_ok=True) + + config_states = [] + for filename in os.listdir(config_states_dir): + if filename.endswith(".json"): + path = os.path.join(config_states_dir, filename) + with open(path, "r", encoding="utf-8") as f: + j = json.load(f) + j["filepath"] = path + config_states.append(j) + + config_states = sorted(config_states, key=lambda cs: cs["created_at"], reverse=True) + + for cs in config_states: + timestamp = time.asctime(time.gmtime(cs["created_at"])) + name = cs.get("name", "Config") + full_name = f"{name}: {timestamp}" + all_config_states[full_name] = cs + + return all_config_states + + +def get_webui_config(): + webui_repo = None + + try: + if os.path.exists(os.path.join(script_path, ".git")): + webui_repo = git.Repo(script_path) + except Exception: + print(f"Error reading webui git info from {script_path}:", file=sys.stderr) + print(traceback.format_exc(), file=sys.stderr) + + webui_remote = None + webui_commit_hash = None + webui_commit_date = None + webui_branch = None + if webui_repo and not webui_repo.bare: + try: + webui_remote = next(webui_repo.remote().urls, None) + head = webui_repo.head.commit + webui_commit_date = webui_repo.head.commit.committed_date + webui_commit_hash = head.hexsha + webui_branch = webui_repo.active_branch.name + + except Exception: + webui_remote = None + + return { + "remote": webui_remote, + "commit_hash": webui_commit_hash, + "commit_date": webui_commit_date, + "branch": webui_branch, + } + + +def get_extension_config(): + ext_config = {} + + for ext in extensions.extensions: + ext.read_info_from_repo() + + entry = { + "name": ext.name, + "path": ext.path, + "enabled": ext.enabled, + "is_builtin": ext.is_builtin, + "remote": ext.remote, + "commit_hash": ext.commit_hash, + "commit_date": ext.commit_date, + "branch": ext.branch, + "have_info_from_repo": ext.have_info_from_repo + } + + ext_config[ext.name] = entry + + return ext_config + + +def get_config(): + creation_time = datetime.now().timestamp() + webui_config = get_webui_config() + ext_config = get_extension_config() + + return { + "created_at": creation_time, + "webui": webui_config, + "extensions": ext_config + } + + +def restore_webui_config(config): + print("* Restoring webui state...") + + if "webui" not in config: + print("Error: No webui data saved to config") + return + + webui_config = config["webui"] + + if "commit_hash" not in webui_config: + print("Error: No commit saved to webui config") + return + + webui_commit_hash = webui_config.get("commit_hash", None) + webui_repo = None + + try: + if os.path.exists(os.path.join(script_path, ".git")): + webui_repo = git.Repo(script_path) + except Exception: + print(f"Error reading webui git info from {script_path}:", file=sys.stderr) + print(traceback.format_exc(), file=sys.stderr) + return + + try: + webui_repo.git.fetch(all=True) + webui_repo.git.reset(webui_commit_hash, hard=True) + print(f"* Restored webui to commit {webui_commit_hash}.") + except Exception: + print(f"Error restoring webui to commit {webui_commit_hash}:", file=sys.stderr) + print(traceback.format_exc(), file=sys.stderr) + + +def restore_extension_config(config): + print("* Restoring extension state...") + + if "extensions" not in config: + print("Error: No extension data saved to config") + return + + ext_config = config["extensions"] + + results = [] + disabled = [] + + for ext in tqdm.tqdm(extensions.extensions): + if ext.is_builtin: + continue + + ext.read_info_from_repo() + current_commit = ext.commit_hash + + if ext.name not in ext_config: + ext.disabled = True + disabled.append(ext.name) + results.append((ext, current_commit[:8], False, "Saved extension state not found in config, marking as disabled")) + continue + + entry = ext_config[ext.name] + + if "commit_hash" in entry and entry["commit_hash"]: + try: + ext.fetch_and_reset_hard(entry["commit_hash"]) + ext.read_info_from_repo() + if current_commit != entry["commit_hash"]: + results.append((ext, current_commit[:8], True, entry["commit_hash"][:8])) + except Exception as ex: + results.append((ext, current_commit[:8], False, ex)) + else: + results.append((ext, current_commit[:8], False, "No commit hash found in config")) + + if not entry.get("enabled", False): + ext.disabled = True + disabled.append(ext.name) + else: + ext.disabled = False + + shared.opts.disabled_extensions = disabled + shared.opts.save(shared.config_filename) + + print("* Finished restoring extensions. Results:") + for ext, prev_commit, success, result in results: + if success: + print(f" + {ext.name}: {prev_commit} -> {result}") + else: + print(f" ! {ext.name}: FAILURE ({result})") diff --git a/modules/deepbooru.py b/modules/deepbooru.py new file mode 100644 index 0000000000000000000000000000000000000000..547e1b4c67aeb75a06c9991f957f51b0ef6fdd0f --- /dev/null +++ b/modules/deepbooru.py @@ -0,0 +1,98 @@ +import os +import re + +import torch +import numpy as np + +from modules import modelloader, paths, deepbooru_model, devices, images, shared + +re_special = re.compile(r'([\\()])') + + +class DeepDanbooru: + def __init__(self): + self.model = None + + def load(self): + if self.model is not None: + return + + files = modelloader.load_models( + model_path=os.path.join(paths.models_path, "torch_deepdanbooru"), + model_url='https://github.com/AUTOMATIC1111/TorchDeepDanbooru/releases/download/v1/model-resnet_custom_v3.pt', + ext_filter=[".pt"], + download_name='model-resnet_custom_v3.pt', + ) + + self.model = deepbooru_model.DeepDanbooruModel() + self.model.load_state_dict(torch.load(files[0], map_location="cpu")) + + self.model.eval() + self.model.to(devices.cpu, devices.dtype) + + def start(self): + self.load() + self.model.to(devices.device) + + def stop(self): + if not shared.opts.interrogate_keep_models_in_memory: + self.model.to(devices.cpu) + devices.torch_gc() + + def tag(self, pil_image): + self.start() + res = self.tag_multi(pil_image) + self.stop() + + return res + + def tag_multi(self, pil_image, force_disable_ranks=False): + threshold = shared.opts.interrogate_deepbooru_score_threshold + use_spaces = shared.opts.deepbooru_use_spaces + use_escape = shared.opts.deepbooru_escape + alpha_sort = shared.opts.deepbooru_sort_alpha + include_ranks = shared.opts.interrogate_return_ranks and not force_disable_ranks + + pic = images.resize_image(2, pil_image.convert("RGB"), 512, 512) + a = np.expand_dims(np.array(pic, dtype=np.float32), 0) / 255 + + with torch.no_grad(), devices.autocast(): + x = torch.from_numpy(a).to(devices.device) + y = self.model(x)[0].detach().cpu().numpy() + + probability_dict = {} + + for tag, probability in zip(self.model.tags, y): + if probability < threshold: + continue + + if tag.startswith("rating:"): + continue + + probability_dict[tag] = probability + + if alpha_sort: + tags = sorted(probability_dict) + else: + tags = [tag for tag, _ in sorted(probability_dict.items(), key=lambda x: -x[1])] + + res = [] + + filtertags = {x.strip().replace(' ', '_') for x in shared.opts.deepbooru_filter_tags.split(",")} + + for tag in [x for x in tags if x not in filtertags]: + probability = probability_dict[tag] + tag_outformat = tag + if use_spaces: + tag_outformat = tag_outformat.replace('_', ' ') + if use_escape: + tag_outformat = re.sub(re_special, r'\\\1', tag_outformat) + if include_ranks: + tag_outformat = f"({tag_outformat}:{probability:.3f})" + + res.append(tag_outformat) + + return ", ".join(res) + + +model = DeepDanbooru() diff --git a/modules/deepbooru_model.py b/modules/deepbooru_model.py new file mode 100644 index 0000000000000000000000000000000000000000..7a53884624e96284c35214ce02b8a2891d92c3e8 --- /dev/null +++ b/modules/deepbooru_model.py @@ -0,0 +1,678 @@ +import torch +import torch.nn as nn +import torch.nn.functional as F + +from modules import devices + +# see https://github.com/AUTOMATIC1111/TorchDeepDanbooru for more + + +class DeepDanbooruModel(nn.Module): + def __init__(self): + super(DeepDanbooruModel, self).__init__() + + self.tags = [] + + self.n_Conv_0 = nn.Conv2d(kernel_size=(7, 7), in_channels=3, out_channels=64, stride=(2, 2)) + self.n_MaxPool_0 = nn.MaxPool2d(kernel_size=(3, 3), stride=(2, 2)) + self.n_Conv_1 = nn.Conv2d(kernel_size=(1, 1), in_channels=64, out_channels=256) + self.n_Conv_2 = nn.Conv2d(kernel_size=(1, 1), in_channels=64, out_channels=64) + self.n_Conv_3 = nn.Conv2d(kernel_size=(3, 3), in_channels=64, out_channels=64) + self.n_Conv_4 = nn.Conv2d(kernel_size=(1, 1), in_channels=64, out_channels=256) + self.n_Conv_5 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=64) + self.n_Conv_6 = nn.Conv2d(kernel_size=(3, 3), in_channels=64, out_channels=64) + self.n_Conv_7 = nn.Conv2d(kernel_size=(1, 1), in_channels=64, out_channels=256) + self.n_Conv_8 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=64) + self.n_Conv_9 = nn.Conv2d(kernel_size=(3, 3), in_channels=64, out_channels=64) + self.n_Conv_10 = nn.Conv2d(kernel_size=(1, 1), in_channels=64, out_channels=256) + self.n_Conv_11 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=512, stride=(2, 2)) + self.n_Conv_12 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=128) + self.n_Conv_13 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128, stride=(2, 2)) + self.n_Conv_14 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512) + self.n_Conv_15 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128) + self.n_Conv_16 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128) + self.n_Conv_17 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512) + self.n_Conv_18 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128) + self.n_Conv_19 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128) + self.n_Conv_20 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512) + self.n_Conv_21 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128) + self.n_Conv_22 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128) + self.n_Conv_23 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512) + self.n_Conv_24 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128) + self.n_Conv_25 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128) + self.n_Conv_26 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512) + self.n_Conv_27 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128) + self.n_Conv_28 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128) + self.n_Conv_29 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512) + self.n_Conv_30 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128) + self.n_Conv_31 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128) + self.n_Conv_32 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512) + self.n_Conv_33 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=128) + self.n_Conv_34 = nn.Conv2d(kernel_size=(3, 3), in_channels=128, out_channels=128) + self.n_Conv_35 = nn.Conv2d(kernel_size=(1, 1), in_channels=128, out_channels=512) + self.n_Conv_36 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=1024, stride=(2, 2)) + self.n_Conv_37 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=256) + self.n_Conv_38 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256, stride=(2, 2)) + self.n_Conv_39 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_40 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_41 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_42 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_43 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_44 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_45 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_46 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_47 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_48 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_49 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_50 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_51 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_52 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_53 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_54 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_55 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_56 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_57 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_58 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_59 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_60 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_61 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_62 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_63 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_64 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_65 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_66 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_67 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_68 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_69 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_70 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_71 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_72 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_73 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_74 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_75 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_76 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_77 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_78 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_79 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_80 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_81 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_82 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_83 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_84 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_85 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_86 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_87 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_88 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_89 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_90 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_91 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_92 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_93 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_94 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_95 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_96 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_97 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_98 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256, stride=(2, 2)) + self.n_Conv_99 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_100 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=1024, stride=(2, 2)) + self.n_Conv_101 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_102 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_103 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_104 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_105 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_106 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_107 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_108 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_109 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_110 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_111 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_112 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_113 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_114 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_115 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_116 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_117 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_118 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_119 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_120 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_121 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_122 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_123 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_124 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_125 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_126 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_127 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_128 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_129 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_130 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_131 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_132 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_133 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_134 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_135 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_136 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_137 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_138 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_139 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_140 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_141 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_142 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_143 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_144 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_145 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_146 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_147 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_148 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_149 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_150 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_151 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_152 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_153 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_154 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_155 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=256) + self.n_Conv_156 = nn.Conv2d(kernel_size=(3, 3), in_channels=256, out_channels=256) + self.n_Conv_157 = nn.Conv2d(kernel_size=(1, 1), in_channels=256, out_channels=1024) + self.n_Conv_158 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=2048, stride=(2, 2)) + self.n_Conv_159 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=512) + self.n_Conv_160 = nn.Conv2d(kernel_size=(3, 3), in_channels=512, out_channels=512, stride=(2, 2)) + self.n_Conv_161 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=2048) + self.n_Conv_162 = nn.Conv2d(kernel_size=(1, 1), in_channels=2048, out_channels=512) + self.n_Conv_163 = nn.Conv2d(kernel_size=(3, 3), in_channels=512, out_channels=512) + self.n_Conv_164 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=2048) + self.n_Conv_165 = nn.Conv2d(kernel_size=(1, 1), in_channels=2048, out_channels=512) + self.n_Conv_166 = nn.Conv2d(kernel_size=(3, 3), in_channels=512, out_channels=512) + self.n_Conv_167 = nn.Conv2d(kernel_size=(1, 1), in_channels=512, out_channels=2048) + self.n_Conv_168 = nn.Conv2d(kernel_size=(1, 1), in_channels=2048, out_channels=4096, stride=(2, 2)) + self.n_Conv_169 = nn.Conv2d(kernel_size=(1, 1), in_channels=2048, out_channels=1024) + self.n_Conv_170 = nn.Conv2d(kernel_size=(3, 3), in_channels=1024, out_channels=1024, stride=(2, 2)) + self.n_Conv_171 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=4096) + self.n_Conv_172 = nn.Conv2d(kernel_size=(1, 1), in_channels=4096, out_channels=1024) + self.n_Conv_173 = nn.Conv2d(kernel_size=(3, 3), in_channels=1024, out_channels=1024) + self.n_Conv_174 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=4096) + self.n_Conv_175 = nn.Conv2d(kernel_size=(1, 1), in_channels=4096, out_channels=1024) + self.n_Conv_176 = nn.Conv2d(kernel_size=(3, 3), in_channels=1024, out_channels=1024) + self.n_Conv_177 = nn.Conv2d(kernel_size=(1, 1), in_channels=1024, out_channels=4096) + self.n_Conv_178 = nn.Conv2d(kernel_size=(1, 1), in_channels=4096, out_channels=9176, bias=False) + + def forward(self, *inputs): + t_358, = inputs + t_359 = t_358.permute(*[0, 3, 1, 2]) + t_359_padded = F.pad(t_359, [2, 3, 2, 3], value=0) + t_360 = self.n_Conv_0(t_359_padded.to(self.n_Conv_0.bias.dtype) if devices.unet_needs_upcast else t_359_padded) + t_361 = F.relu(t_360) + t_361 = F.pad(t_361, [0, 1, 0, 1], value=float('-inf')) + t_362 = self.n_MaxPool_0(t_361) + t_363 = self.n_Conv_1(t_362) + t_364 = self.n_Conv_2(t_362) + t_365 = F.relu(t_364) + t_365_padded = F.pad(t_365, [1, 1, 1, 1], value=0) + t_366 = self.n_Conv_3(t_365_padded) + t_367 = F.relu(t_366) + t_368 = self.n_Conv_4(t_367) + t_369 = torch.add(t_368, t_363) + t_370 = F.relu(t_369) + t_371 = self.n_Conv_5(t_370) + t_372 = F.relu(t_371) + t_372_padded = F.pad(t_372, [1, 1, 1, 1], value=0) + t_373 = self.n_Conv_6(t_372_padded) + t_374 = F.relu(t_373) + t_375 = self.n_Conv_7(t_374) + t_376 = torch.add(t_375, t_370) + t_377 = F.relu(t_376) + t_378 = self.n_Conv_8(t_377) + t_379 = F.relu(t_378) + t_379_padded = F.pad(t_379, [1, 1, 1, 1], value=0) + t_380 = self.n_Conv_9(t_379_padded) + t_381 = F.relu(t_380) + t_382 = self.n_Conv_10(t_381) + t_383 = torch.add(t_382, t_377) + t_384 = F.relu(t_383) + t_385 = self.n_Conv_11(t_384) + t_386 = self.n_Conv_12(t_384) + t_387 = F.relu(t_386) + t_387_padded = F.pad(t_387, [0, 1, 0, 1], value=0) + t_388 = self.n_Conv_13(t_387_padded) + t_389 = F.relu(t_388) + t_390 = self.n_Conv_14(t_389) + t_391 = torch.add(t_390, t_385) + t_392 = F.relu(t_391) + t_393 = self.n_Conv_15(t_392) + t_394 = F.relu(t_393) + t_394_padded = F.pad(t_394, [1, 1, 1, 1], value=0) + t_395 = self.n_Conv_16(t_394_padded) + t_396 = F.relu(t_395) + t_397 = self.n_Conv_17(t_396) + t_398 = torch.add(t_397, t_392) + t_399 = F.relu(t_398) + t_400 = self.n_Conv_18(t_399) + t_401 = F.relu(t_400) + t_401_padded = F.pad(t_401, [1, 1, 1, 1], value=0) + t_402 = self.n_Conv_19(t_401_padded) + t_403 = F.relu(t_402) + t_404 = self.n_Conv_20(t_403) + t_405 = torch.add(t_404, t_399) + t_406 = F.relu(t_405) + t_407 = self.n_Conv_21(t_406) + t_408 = F.relu(t_407) + t_408_padded = F.pad(t_408, [1, 1, 1, 1], value=0) + t_409 = self.n_Conv_22(t_408_padded) + t_410 = F.relu(t_409) + t_411 = self.n_Conv_23(t_410) + t_412 = torch.add(t_411, t_406) + t_413 = F.relu(t_412) + t_414 = self.n_Conv_24(t_413) + t_415 = F.relu(t_414) + t_415_padded = F.pad(t_415, [1, 1, 1, 1], value=0) + t_416 = self.n_Conv_25(t_415_padded) + t_417 = F.relu(t_416) + t_418 = self.n_Conv_26(t_417) + t_419 = torch.add(t_418, t_413) + t_420 = F.relu(t_419) + t_421 = self.n_Conv_27(t_420) + t_422 = F.relu(t_421) + t_422_padded = F.pad(t_422, [1, 1, 1, 1], value=0) + t_423 = self.n_Conv_28(t_422_padded) + t_424 = F.relu(t_423) + t_425 = self.n_Conv_29(t_424) + t_426 = torch.add(t_425, t_420) + t_427 = F.relu(t_426) + t_428 = self.n_Conv_30(t_427) + t_429 = F.relu(t_428) + t_429_padded = F.pad(t_429, [1, 1, 1, 1], value=0) + t_430 = self.n_Conv_31(t_429_padded) + t_431 = F.relu(t_430) + t_432 = self.n_Conv_32(t_431) + t_433 = torch.add(t_432, t_427) + t_434 = F.relu(t_433) + t_435 = self.n_Conv_33(t_434) + t_436 = F.relu(t_435) + t_436_padded = F.pad(t_436, [1, 1, 1, 1], value=0) + t_437 = self.n_Conv_34(t_436_padded) + t_438 = F.relu(t_437) + t_439 = self.n_Conv_35(t_438) + t_440 = torch.add(t_439, t_434) + t_441 = F.relu(t_440) + t_442 = self.n_Conv_36(t_441) + t_443 = self.n_Conv_37(t_441) + t_444 = F.relu(t_443) + t_444_padded = F.pad(t_444, [0, 1, 0, 1], value=0) + t_445 = self.n_Conv_38(t_444_padded) + t_446 = F.relu(t_445) + t_447 = self.n_Conv_39(t_446) + t_448 = torch.add(t_447, t_442) + t_449 = F.relu(t_448) + t_450 = self.n_Conv_40(t_449) + t_451 = F.relu(t_450) + t_451_padded = F.pad(t_451, [1, 1, 1, 1], value=0) + t_452 = self.n_Conv_41(t_451_padded) + t_453 = F.relu(t_452) + t_454 = self.n_Conv_42(t_453) + t_455 = torch.add(t_454, t_449) + t_456 = F.relu(t_455) + t_457 = self.n_Conv_43(t_456) + t_458 = F.relu(t_457) + t_458_padded = F.pad(t_458, [1, 1, 1, 1], value=0) + t_459 = self.n_Conv_44(t_458_padded) + t_460 = F.relu(t_459) + t_461 = self.n_Conv_45(t_460) + t_462 = torch.add(t_461, t_456) + t_463 = F.relu(t_462) + t_464 = self.n_Conv_46(t_463) + t_465 = F.relu(t_464) + t_465_padded = F.pad(t_465, [1, 1, 1, 1], value=0) + t_466 = self.n_Conv_47(t_465_padded) + t_467 = F.relu(t_466) + t_468 = self.n_Conv_48(t_467) + t_469 = torch.add(t_468, t_463) + t_470 = F.relu(t_469) + t_471 = self.n_Conv_49(t_470) + t_472 = F.relu(t_471) + t_472_padded = F.pad(t_472, [1, 1, 1, 1], value=0) + t_473 = self.n_Conv_50(t_472_padded) + t_474 = F.relu(t_473) + t_475 = self.n_Conv_51(t_474) + t_476 = torch.add(t_475, t_470) + t_477 = F.relu(t_476) + t_478 = self.n_Conv_52(t_477) + t_479 = F.relu(t_478) + t_479_padded = F.pad(t_479, [1, 1, 1, 1], value=0) + t_480 = self.n_Conv_53(t_479_padded) + t_481 = F.relu(t_480) + t_482 = self.n_Conv_54(t_481) + t_483 = torch.add(t_482, t_477) + t_484 = F.relu(t_483) + t_485 = self.n_Conv_55(t_484) + t_486 = F.relu(t_485) + t_486_padded = F.pad(t_486, [1, 1, 1, 1], value=0) + t_487 = self.n_Conv_56(t_486_padded) + t_488 = F.relu(t_487) + t_489 = self.n_Conv_57(t_488) + t_490 = torch.add(t_489, t_484) + t_491 = F.relu(t_490) + t_492 = self.n_Conv_58(t_491) + t_493 = F.relu(t_492) + t_493_padded = F.pad(t_493, [1, 1, 1, 1], value=0) + t_494 = self.n_Conv_59(t_493_padded) + t_495 = F.relu(t_494) + t_496 = self.n_Conv_60(t_495) + t_497 = torch.add(t_496, t_491) + t_498 = F.relu(t_497) + t_499 = self.n_Conv_61(t_498) + t_500 = F.relu(t_499) + t_500_padded = F.pad(t_500, [1, 1, 1, 1], value=0) + t_501 = self.n_Conv_62(t_500_padded) + t_502 = F.relu(t_501) + t_503 = self.n_Conv_63(t_502) + t_504 = torch.add(t_503, t_498) + t_505 = F.relu(t_504) + t_506 = self.n_Conv_64(t_505) + t_507 = F.relu(t_506) + t_507_padded = F.pad(t_507, [1, 1, 1, 1], value=0) + t_508 = self.n_Conv_65(t_507_padded) + t_509 = F.relu(t_508) + t_510 = self.n_Conv_66(t_509) + t_511 = torch.add(t_510, t_505) + t_512 = F.relu(t_511) + t_513 = self.n_Conv_67(t_512) + t_514 = F.relu(t_513) + t_514_padded = F.pad(t_514, [1, 1, 1, 1], value=0) + t_515 = self.n_Conv_68(t_514_padded) + t_516 = F.relu(t_515) + t_517 = self.n_Conv_69(t_516) + t_518 = torch.add(t_517, t_512) + t_519 = F.relu(t_518) + t_520 = self.n_Conv_70(t_519) + t_521 = F.relu(t_520) + t_521_padded = F.pad(t_521, [1, 1, 1, 1], value=0) + t_522 = self.n_Conv_71(t_521_padded) + t_523 = F.relu(t_522) + t_524 = self.n_Conv_72(t_523) + t_525 = torch.add(t_524, t_519) + t_526 = F.relu(t_525) + t_527 = self.n_Conv_73(t_526) + t_528 = F.relu(t_527) + t_528_padded = F.pad(t_528, [1, 1, 1, 1], value=0) + t_529 = self.n_Conv_74(t_528_padded) + t_530 = F.relu(t_529) + t_531 = self.n_Conv_75(t_530) + t_532 = torch.add(t_531, t_526) + t_533 = F.relu(t_532) + t_534 = self.n_Conv_76(t_533) + t_535 = F.relu(t_534) + t_535_padded = F.pad(t_535, [1, 1, 1, 1], value=0) + t_536 = self.n_Conv_77(t_535_padded) + t_537 = F.relu(t_536) + t_538 = self.n_Conv_78(t_537) + t_539 = torch.add(t_538, t_533) + t_540 = F.relu(t_539) + t_541 = self.n_Conv_79(t_540) + t_542 = F.relu(t_541) + t_542_padded = F.pad(t_542, [1, 1, 1, 1], value=0) + t_543 = self.n_Conv_80(t_542_padded) + t_544 = F.relu(t_543) + t_545 = self.n_Conv_81(t_544) + t_546 = torch.add(t_545, t_540) + t_547 = F.relu(t_546) + t_548 = self.n_Conv_82(t_547) + t_549 = F.relu(t_548) + t_549_padded = F.pad(t_549, [1, 1, 1, 1], value=0) + t_550 = self.n_Conv_83(t_549_padded) + t_551 = F.relu(t_550) + t_552 = self.n_Conv_84(t_551) + t_553 = torch.add(t_552, t_547) + t_554 = F.relu(t_553) + t_555 = self.n_Conv_85(t_554) + t_556 = F.relu(t_555) + t_556_padded = F.pad(t_556, [1, 1, 1, 1], value=0) + t_557 = self.n_Conv_86(t_556_padded) + t_558 = F.relu(t_557) + t_559 = self.n_Conv_87(t_558) + t_560 = torch.add(t_559, t_554) + t_561 = F.relu(t_560) + t_562 = self.n_Conv_88(t_561) + t_563 = F.relu(t_562) + t_563_padded = F.pad(t_563, [1, 1, 1, 1], value=0) + t_564 = self.n_Conv_89(t_563_padded) + t_565 = F.relu(t_564) + t_566 = self.n_Conv_90(t_565) + t_567 = torch.add(t_566, t_561) + t_568 = F.relu(t_567) + t_569 = self.n_Conv_91(t_568) + t_570 = F.relu(t_569) + t_570_padded = F.pad(t_570, [1, 1, 1, 1], value=0) + t_571 = self.n_Conv_92(t_570_padded) + t_572 = F.relu(t_571) + t_573 = self.n_Conv_93(t_572) + t_574 = torch.add(t_573, t_568) + t_575 = F.relu(t_574) + t_576 = self.n_Conv_94(t_575) + t_577 = F.relu(t_576) + t_577_padded = F.pad(t_577, [1, 1, 1, 1], value=0) + t_578 = self.n_Conv_95(t_577_padded) + t_579 = F.relu(t_578) + t_580 = self.n_Conv_96(t_579) + t_581 = torch.add(t_580, t_575) + t_582 = F.relu(t_581) + t_583 = self.n_Conv_97(t_582) + t_584 = F.relu(t_583) + t_584_padded = F.pad(t_584, [0, 1, 0, 1], value=0) + t_585 = self.n_Conv_98(t_584_padded) + t_586 = F.relu(t_585) + t_587 = self.n_Conv_99(t_586) + t_588 = self.n_Conv_100(t_582) + t_589 = torch.add(t_587, t_588) + t_590 = F.relu(t_589) + t_591 = self.n_Conv_101(t_590) + t_592 = F.relu(t_591) + t_592_padded = F.pad(t_592, [1, 1, 1, 1], value=0) + t_593 = self.n_Conv_102(t_592_padded) + t_594 = F.relu(t_593) + t_595 = self.n_Conv_103(t_594) + t_596 = torch.add(t_595, t_590) + t_597 = F.relu(t_596) + t_598 = self.n_Conv_104(t_597) + t_599 = F.relu(t_598) + t_599_padded = F.pad(t_599, [1, 1, 1, 1], value=0) + t_600 = self.n_Conv_105(t_599_padded) + t_601 = F.relu(t_600) + t_602 = self.n_Conv_106(t_601) + t_603 = torch.add(t_602, t_597) + t_604 = F.relu(t_603) + t_605 = self.n_Conv_107(t_604) + t_606 = F.relu(t_605) + t_606_padded = F.pad(t_606, [1, 1, 1, 1], value=0) + t_607 = self.n_Conv_108(t_606_padded) + t_608 = F.relu(t_607) + t_609 = self.n_Conv_109(t_608) + t_610 = torch.add(t_609, t_604) + t_611 = F.relu(t_610) + t_612 = self.n_Conv_110(t_611) + t_613 = F.relu(t_612) + t_613_padded = F.pad(t_613, [1, 1, 1, 1], value=0) + t_614 = self.n_Conv_111(t_613_padded) + t_615 = F.relu(t_614) + t_616 = self.n_Conv_112(t_615) + t_617 = torch.add(t_616, t_611) + t_618 = F.relu(t_617) + t_619 = self.n_Conv_113(t_618) + t_620 = F.relu(t_619) + t_620_padded = F.pad(t_620, [1, 1, 1, 1], value=0) + t_621 = self.n_Conv_114(t_620_padded) + t_622 = F.relu(t_621) + t_623 = self.n_Conv_115(t_622) + t_624 = torch.add(t_623, t_618) + t_625 = F.relu(t_624) + t_626 = self.n_Conv_116(t_625) + t_627 = F.relu(t_626) + t_627_padded = F.pad(t_627, [1, 1, 1, 1], value=0) + t_628 = self.n_Conv_117(t_627_padded) + t_629 = F.relu(t_628) + t_630 = self.n_Conv_118(t_629) + t_631 = torch.add(t_630, t_625) + t_632 = F.relu(t_631) + t_633 = self.n_Conv_119(t_632) + t_634 = F.relu(t_633) + t_634_padded = F.pad(t_634, [1, 1, 1, 1], value=0) + t_635 = self.n_Conv_120(t_634_padded) + t_636 = F.relu(t_635) + t_637 = self.n_Conv_121(t_636) + t_638 = torch.add(t_637, t_632) + t_639 = F.relu(t_638) + t_640 = self.n_Conv_122(t_639) + t_641 = F.relu(t_640) + t_641_padded = F.pad(t_641, [1, 1, 1, 1], value=0) + t_642 = self.n_Conv_123(t_641_padded) + t_643 = F.relu(t_642) + t_644 = self.n_Conv_124(t_643) + t_645 = torch.add(t_644, t_639) + t_646 = F.relu(t_645) + t_647 = self.n_Conv_125(t_646) + t_648 = F.relu(t_647) + t_648_padded = F.pad(t_648, [1, 1, 1, 1], value=0) + t_649 = self.n_Conv_126(t_648_padded) + t_650 = F.relu(t_649) + t_651 = self.n_Conv_127(t_650) + t_652 = torch.add(t_651, t_646) + t_653 = F.relu(t_652) + t_654 = self.n_Conv_128(t_653) + t_655 = F.relu(t_654) + t_655_padded = F.pad(t_655, [1, 1, 1, 1], value=0) + t_656 = self.n_Conv_129(t_655_padded) + t_657 = F.relu(t_656) + t_658 = self.n_Conv_130(t_657) + t_659 = torch.add(t_658, t_653) + t_660 = F.relu(t_659) + t_661 = self.n_Conv_131(t_660) + t_662 = F.relu(t_661) + t_662_padded = F.pad(t_662, [1, 1, 1, 1], value=0) + t_663 = self.n_Conv_132(t_662_padded) + t_664 = F.relu(t_663) + t_665 = self.n_Conv_133(t_664) + t_666 = torch.add(t_665, t_660) + t_667 = F.relu(t_666) + t_668 = self.n_Conv_134(t_667) + t_669 = F.relu(t_668) + t_669_padded = F.pad(t_669, [1, 1, 1, 1], value=0) + t_670 = self.n_Conv_135(t_669_padded) + t_671 = F.relu(t_670) + t_672 = self.n_Conv_136(t_671) + t_673 = torch.add(t_672, t_667) + t_674 = F.relu(t_673) + t_675 = self.n_Conv_137(t_674) + t_676 = F.relu(t_675) + t_676_padded = F.pad(t_676, [1, 1, 1, 1], value=0) + t_677 = self.n_Conv_138(t_676_padded) + t_678 = F.relu(t_677) + t_679 = self.n_Conv_139(t_678) + t_680 = torch.add(t_679, t_674) + t_681 = F.relu(t_680) + t_682 = self.n_Conv_140(t_681) + t_683 = F.relu(t_682) + t_683_padded = F.pad(t_683, [1, 1, 1, 1], value=0) + t_684 = self.n_Conv_141(t_683_padded) + t_685 = F.relu(t_684) + t_686 = self.n_Conv_142(t_685) + t_687 = torch.add(t_686, t_681) + t_688 = F.relu(t_687) + t_689 = self.n_Conv_143(t_688) + t_690 = F.relu(t_689) + t_690_padded = F.pad(t_690, [1, 1, 1, 1], value=0) + t_691 = self.n_Conv_144(t_690_padded) + t_692 = F.relu(t_691) + t_693 = self.n_Conv_145(t_692) + t_694 = torch.add(t_693, t_688) + t_695 = F.relu(t_694) + t_696 = self.n_Conv_146(t_695) + t_697 = F.relu(t_696) + t_697_padded = F.pad(t_697, [1, 1, 1, 1], value=0) + t_698 = self.n_Conv_147(t_697_padded) + t_699 = F.relu(t_698) + t_700 = self.n_Conv_148(t_699) + t_701 = torch.add(t_700, t_695) + t_702 = F.relu(t_701) + t_703 = self.n_Conv_149(t_702) + t_704 = F.relu(t_703) + t_704_padded = F.pad(t_704, [1, 1, 1, 1], value=0) + t_705 = self.n_Conv_150(t_704_padded) + t_706 = F.relu(t_705) + t_707 = self.n_Conv_151(t_706) + t_708 = torch.add(t_707, t_702) + t_709 = F.relu(t_708) + t_710 = self.n_Conv_152(t_709) + t_711 = F.relu(t_710) + t_711_padded = F.pad(t_711, [1, 1, 1, 1], value=0) + t_712 = self.n_Conv_153(t_711_padded) + t_713 = F.relu(t_712) + t_714 = self.n_Conv_154(t_713) + t_715 = torch.add(t_714, t_709) + t_716 = F.relu(t_715) + t_717 = self.n_Conv_155(t_716) + t_718 = F.relu(t_717) + t_718_padded = F.pad(t_718, [1, 1, 1, 1], value=0) + t_719 = self.n_Conv_156(t_718_padded) + t_720 = F.relu(t_719) + t_721 = self.n_Conv_157(t_720) + t_722 = torch.add(t_721, t_716) + t_723 = F.relu(t_722) + t_724 = self.n_Conv_158(t_723) + t_725 = self.n_Conv_159(t_723) + t_726 = F.relu(t_725) + t_726_padded = F.pad(t_726, [0, 1, 0, 1], value=0) + t_727 = self.n_Conv_160(t_726_padded) + t_728 = F.relu(t_727) + t_729 = self.n_Conv_161(t_728) + t_730 = torch.add(t_729, t_724) + t_731 = F.relu(t_730) + t_732 = self.n_Conv_162(t_731) + t_733 = F.relu(t_732) + t_733_padded = F.pad(t_733, [1, 1, 1, 1], value=0) + t_734 = self.n_Conv_163(t_733_padded) + t_735 = F.relu(t_734) + t_736 = self.n_Conv_164(t_735) + t_737 = torch.add(t_736, t_731) + t_738 = F.relu(t_737) + t_739 = self.n_Conv_165(t_738) + t_740 = F.relu(t_739) + t_740_padded = F.pad(t_740, [1, 1, 1, 1], value=0) + t_741 = self.n_Conv_166(t_740_padded) + t_742 = F.relu(t_741) + t_743 = self.n_Conv_167(t_742) + t_744 = torch.add(t_743, t_738) + t_745 = F.relu(t_744) + t_746 = self.n_Conv_168(t_745) + t_747 = self.n_Conv_169(t_745) + t_748 = F.relu(t_747) + t_748_padded = F.pad(t_748, [0, 1, 0, 1], value=0) + t_749 = self.n_Conv_170(t_748_padded) + t_750 = F.relu(t_749) + t_751 = self.n_Conv_171(t_750) + t_752 = torch.add(t_751, t_746) + t_753 = F.relu(t_752) + t_754 = self.n_Conv_172(t_753) + t_755 = F.relu(t_754) + t_755_padded = F.pad(t_755, [1, 1, 1, 1], value=0) + t_756 = self.n_Conv_173(t_755_padded) + t_757 = F.relu(t_756) + t_758 = self.n_Conv_174(t_757) + t_759 = torch.add(t_758, t_753) + t_760 = F.relu(t_759) + t_761 = self.n_Conv_175(t_760) + t_762 = F.relu(t_761) + t_762_padded = F.pad(t_762, [1, 1, 1, 1], value=0) + t_763 = self.n_Conv_176(t_762_padded) + t_764 = F.relu(t_763) + t_765 = self.n_Conv_177(t_764) + t_766 = torch.add(t_765, t_760) + t_767 = F.relu(t_766) + t_768 = self.n_Conv_178(t_767) + t_769 = F.avg_pool2d(t_768, kernel_size=t_768.shape[-2:]) + t_770 = torch.squeeze(t_769, 3) + t_770 = torch.squeeze(t_770, 2) + t_771 = torch.sigmoid(t_770) + return t_771 + + def load_state_dict(self, state_dict, **kwargs): + self.tags = state_dict.get('tags', []) + + super(DeepDanbooruModel, self).load_state_dict({k: v for k, v in state_dict.items() if k != 'tags'}) + diff --git a/modules/devices.py b/modules/devices.py new file mode 100644 index 0000000000000000000000000000000000000000..d8a34a0fdde27402c45c9d22ea316014b3d6db73 --- /dev/null +++ b/modules/devices.py @@ -0,0 +1,156 @@ +import sys +import contextlib +import torch +from modules import errors + +if sys.platform == "darwin": + from modules import mac_specific + + +def has_mps() -> bool: + if sys.platform != "darwin": + return False + else: + return mac_specific.has_mps + +def extract_device_id(args, name): + for x in range(len(args)): + if name in args[x]: + return args[x + 1] + + return None + + +def get_cuda_device_string(): + from modules import shared + + if shared.cmd_opts.device_id is not None: + return f"cuda:{shared.cmd_opts.device_id}" + + return "cuda" + + +def get_optimal_device_name(): + if torch.cuda.is_available(): + return get_cuda_device_string() + + if has_mps(): + return "mps" + + return "cpu" + + +def get_optimal_device(): + return torch.device(get_optimal_device_name()) + + +def get_device_for(task): + from modules import shared + + if task in shared.cmd_opts.use_cpu: + return cpu + + return get_optimal_device() + + +def torch_gc(): + if torch.cuda.is_available(): + with torch.cuda.device(get_cuda_device_string()): + torch.cuda.empty_cache() + torch.cuda.ipc_collect() + + +def enable_tf32(): + if torch.cuda.is_available(): + + # enabling benchmark option seems to enable a range of cards to do fp16 when they otherwise can't + # see https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/4407 + if any(torch.cuda.get_device_capability(devid) == (7, 5) for devid in range(0, torch.cuda.device_count())): + torch.backends.cudnn.benchmark = True + + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + + + +errors.run(enable_tf32, "Enabling TF32") + +cpu = torch.device("cpu") +device = device_interrogate = device_gfpgan = device_esrgan = device_codeformer = None +dtype = torch.float16 +dtype_vae = torch.float16 +dtype_unet = torch.float16 +unet_needs_upcast = False + + +def cond_cast_unet(input): + return input.to(dtype_unet) if unet_needs_upcast else input + + +def cond_cast_float(input): + return input.float() if unet_needs_upcast else input + + +def randn(seed, shape): + from modules.shared import opts + + torch.manual_seed(seed) + if opts.randn_source == "CPU" or device.type == 'mps': + return torch.randn(shape, device=cpu).to(device) + return torch.randn(shape, device=device) + + +def randn_without_seed(shape): + from modules.shared import opts + + if opts.randn_source == "CPU" or device.type == 'mps': + return torch.randn(shape, device=cpu).to(device) + return torch.randn(shape, device=device) + + +def autocast(disable=False): + from modules import shared + + if disable: + return contextlib.nullcontext() + + if dtype == torch.float32 or shared.cmd_opts.precision == "full": + return contextlib.nullcontext() + + return torch.autocast("cuda") + + +def without_autocast(disable=False): + return torch.autocast("cuda", enabled=False) if torch.is_autocast_enabled() and not disable else contextlib.nullcontext() + + +class NansException(Exception): + pass + + +def test_for_nans(x, where): + from modules import shared + + if shared.cmd_opts.disable_nan_check: + return + + if not torch.all(torch.isnan(x)).item(): + return + + if where == "unet": + message = "A tensor with all NaNs was produced in Unet." + + if not shared.cmd_opts.no_half: + message += " This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the \"Upcast cross attention layer to float32\" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this." + + elif where == "vae": + message = "A tensor with all NaNs was produced in VAE." + + if not shared.cmd_opts.no_half and not shared.cmd_opts.no_half_vae: + message += " This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this." + else: + message = "A tensor with all NaNs was produced." + + message += " Use --disable-nan-check commandline argument to disable this check." + + raise NansException(message) diff --git a/modules/errors.py b/modules/errors.py new file mode 100644 index 0000000000000000000000000000000000000000..72c9c44497221eb814b402aa5859a3e6aaeaac00 --- /dev/null +++ b/modules/errors.py @@ -0,0 +1,43 @@ +import sys +import traceback + + +def print_error_explanation(message): + lines = message.strip().split("\n") + max_len = max([len(x) for x in lines]) + + print('=' * max_len, file=sys.stderr) + for line in lines: + print(line, file=sys.stderr) + print('=' * max_len, file=sys.stderr) + + +def display(e: Exception, task): + print(f"{task or 'error'}: {type(e).__name__}", file=sys.stderr) + print(traceback.format_exc(), file=sys.stderr) + + message = str(e) + if "copying a param with shape torch.Size([640, 1024]) from checkpoint, the shape in current model is torch.Size([640, 768])" in message: + print_error_explanation(""" +The most likely cause of this is you are trying to load Stable Diffusion 2.0 model without specifying its config file. +See https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#stable-diffusion-20 for how to solve this. + """) + + +already_displayed = {} + + +def display_once(e: Exception, task): + if task in already_displayed: + return + + display(e, task) + + already_displayed[task] = 1 + + +def run(code, task): + try: + code() + except Exception as e: + display(task, e) diff --git a/modules/esrgan_model.py b/modules/esrgan_model.py new file mode 100644 index 0000000000000000000000000000000000000000..c4caf795122066b57b53c0a36522768b6cf8e4e8 --- /dev/null +++ b/modules/esrgan_model.py @@ -0,0 +1,232 @@ +import os + +import numpy as np +import torch +from PIL import Image +from basicsr.utils.download_util import load_file_from_url + +import modules.esrgan_model_arch as arch +from modules import modelloader, images, devices +from modules.upscaler import Upscaler, UpscalerData +from modules.shared import opts + + + +def mod2normal(state_dict): + # this code is copied from https://github.com/victorca25/iNNfer + if 'conv_first.weight' in state_dict: + crt_net = {} + items = list(state_dict) + + crt_net['model.0.weight'] = state_dict['conv_first.weight'] + crt_net['model.0.bias'] = state_dict['conv_first.bias'] + + for k in items.copy(): + if 'RDB' in k: + ori_k = k.replace('RRDB_trunk.', 'model.1.sub.') + if '.weight' in k: + ori_k = ori_k.replace('.weight', '.0.weight') + elif '.bias' in k: + ori_k = ori_k.replace('.bias', '.0.bias') + crt_net[ori_k] = state_dict[k] + items.remove(k) + + crt_net['model.1.sub.23.weight'] = state_dict['trunk_conv.weight'] + crt_net['model.1.sub.23.bias'] = state_dict['trunk_conv.bias'] + crt_net['model.3.weight'] = state_dict['upconv1.weight'] + crt_net['model.3.bias'] = state_dict['upconv1.bias'] + crt_net['model.6.weight'] = state_dict['upconv2.weight'] + crt_net['model.6.bias'] = state_dict['upconv2.bias'] + crt_net['model.8.weight'] = state_dict['HRconv.weight'] + crt_net['model.8.bias'] = state_dict['HRconv.bias'] + crt_net['model.10.weight'] = state_dict['conv_last.weight'] + crt_net['model.10.bias'] = state_dict['conv_last.bias'] + state_dict = crt_net + return state_dict + + +def resrgan2normal(state_dict, nb=23): + # this code is copied from https://github.com/victorca25/iNNfer + if "conv_first.weight" in state_dict and "body.0.rdb1.conv1.weight" in state_dict: + re8x = 0 + crt_net = {} + items = list(state_dict) + + crt_net['model.0.weight'] = state_dict['conv_first.weight'] + crt_net['model.0.bias'] = state_dict['conv_first.bias'] + + for k in items.copy(): + if "rdb" in k: + ori_k = k.replace('body.', 'model.1.sub.') + ori_k = ori_k.replace('.rdb', '.RDB') + if '.weight' in k: + ori_k = ori_k.replace('.weight', '.0.weight') + elif '.bias' in k: + ori_k = ori_k.replace('.bias', '.0.bias') + crt_net[ori_k] = state_dict[k] + items.remove(k) + + crt_net[f'model.1.sub.{nb}.weight'] = state_dict['conv_body.weight'] + crt_net[f'model.1.sub.{nb}.bias'] = state_dict['conv_body.bias'] + crt_net['model.3.weight'] = state_dict['conv_up1.weight'] + crt_net['model.3.bias'] = state_dict['conv_up1.bias'] + crt_net['model.6.weight'] = state_dict['conv_up2.weight'] + crt_net['model.6.bias'] = state_dict['conv_up2.bias'] + + if 'conv_up3.weight' in state_dict: + # modification supporting: https://github.com/ai-forever/Real-ESRGAN/blob/main/RealESRGAN/rrdbnet_arch.py + re8x = 3 + crt_net['model.9.weight'] = state_dict['conv_up3.weight'] + crt_net['model.9.bias'] = state_dict['conv_up3.bias'] + + crt_net[f'model.{8+re8x}.weight'] = state_dict['conv_hr.weight'] + crt_net[f'model.{8+re8x}.bias'] = state_dict['conv_hr.bias'] + crt_net[f'model.{10+re8x}.weight'] = state_dict['conv_last.weight'] + crt_net[f'model.{10+re8x}.bias'] = state_dict['conv_last.bias'] + + state_dict = crt_net + return state_dict + + +def infer_params(state_dict): + # this code is copied from https://github.com/victorca25/iNNfer + scale2x = 0 + scalemin = 6 + n_uplayer = 0 + plus = False + + for block in list(state_dict): + parts = block.split(".") + n_parts = len(parts) + if n_parts == 5 and parts[2] == "sub": + nb = int(parts[3]) + elif n_parts == 3: + part_num = int(parts[1]) + if (part_num > scalemin + and parts[0] == "model" + and parts[2] == "weight"): + scale2x += 1 + if part_num > n_uplayer: + n_uplayer = part_num + out_nc = state_dict[block].shape[0] + if not plus and "conv1x1" in block: + plus = True + + nf = state_dict["model.0.weight"].shape[0] + in_nc = state_dict["model.0.weight"].shape[1] + out_nc = out_nc + scale = 2 ** scale2x + + return in_nc, out_nc, nf, nb, plus, scale + + +class UpscalerESRGAN(Upscaler): + def __init__(self, dirname): + self.name = "ESRGAN" + self.model_url = "https://github.com/cszn/KAIR/releases/download/v1.0/ESRGAN.pth" + self.model_name = "ESRGAN_4x" + self.scalers = [] + self.user_path = dirname + super().__init__() + model_paths = self.find_models(ext_filter=[".pt", ".pth"]) + scalers = [] + if len(model_paths) == 0: + scaler_data = UpscalerData(self.model_name, self.model_url, self, 4) + scalers.append(scaler_data) + for file in model_paths: + if "http" in file: + name = self.model_name + else: + name = modelloader.friendly_name(file) + + scaler_data = UpscalerData(name, file, self, 4) + self.scalers.append(scaler_data) + + def do_upscale(self, img, selected_model): + model = self.load_model(selected_model) + if model is None: + return img + model.to(devices.device_esrgan) + img = esrgan_upscale(model, img) + return img + + def load_model(self, path: str): + if "http" in path: + filename = load_file_from_url( + url=self.model_url, + model_dir=self.model_download_path, + file_name=f"{self.model_name}.pth", + progress=True, + ) + else: + filename = path + if not os.path.exists(filename) or filename is None: + print(f"Unable to load {self.model_path} from {filename}") + return None + + state_dict = torch.load(filename, map_location='cpu' if devices.device_esrgan.type == 'mps' else None) + + if "params_ema" in state_dict: + state_dict = state_dict["params_ema"] + elif "params" in state_dict: + state_dict = state_dict["params"] + num_conv = 16 if "realesr-animevideov3" in filename else 32 + model = arch.SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=num_conv, upscale=4, act_type='prelu') + model.load_state_dict(state_dict) + model.eval() + return model + + if "body.0.rdb1.conv1.weight" in state_dict and "conv_first.weight" in state_dict: + nb = 6 if "RealESRGAN_x4plus_anime_6B" in filename else 23 + state_dict = resrgan2normal(state_dict, nb) + elif "conv_first.weight" in state_dict: + state_dict = mod2normal(state_dict) + elif "model.0.weight" not in state_dict: + raise Exception("The file is not a recognized ESRGAN model.") + + in_nc, out_nc, nf, nb, plus, mscale = infer_params(state_dict) + + model = arch.RRDBNet(in_nc=in_nc, out_nc=out_nc, nf=nf, nb=nb, upscale=mscale, plus=plus) + model.load_state_dict(state_dict) + model.eval() + + return model + + +def upscale_without_tiling(model, img): + img = np.array(img) + img = img[:, :, ::-1] + img = np.ascontiguousarray(np.transpose(img, (2, 0, 1))) / 255 + img = torch.from_numpy(img).float() + img = img.unsqueeze(0).to(devices.device_esrgan) + with torch.no_grad(): + output = model(img) + output = output.squeeze().float().cpu().clamp_(0, 1).numpy() + output = 255. * np.moveaxis(output, 0, 2) + output = output.astype(np.uint8) + output = output[:, :, ::-1] + return Image.fromarray(output, 'RGB') + + +def esrgan_upscale(model, img): + if opts.ESRGAN_tile == 0: + return upscale_without_tiling(model, img) + + grid = images.split_grid(img, opts.ESRGAN_tile, opts.ESRGAN_tile, opts.ESRGAN_tile_overlap) + newtiles = [] + scale_factor = 1 + + for y, h, row in grid.tiles: + newrow = [] + for tiledata in row: + x, w, tile = tiledata + + output = upscale_without_tiling(model, tile) + scale_factor = output.width // tile.width + + newrow.append([x * scale_factor, w * scale_factor, output]) + newtiles.append([y * scale_factor, h * scale_factor, newrow]) + + newgrid = images.Grid(newtiles, grid.tile_w * scale_factor, grid.tile_h * scale_factor, grid.image_w * scale_factor, grid.image_h * scale_factor, grid.overlap * scale_factor) + output = images.combine_grid(newgrid) + return output diff --git a/modules/esrgan_model_arch.py b/modules/esrgan_model_arch.py new file mode 100644 index 0000000000000000000000000000000000000000..353c70dd867cb894a0ac208f39394280175e4e14 --- /dev/null +++ b/modules/esrgan_model_arch.py @@ -0,0 +1,465 @@ +# this file is adapted from https://github.com/victorca25/iNNfer + +from collections import OrderedDict +import math +import torch +import torch.nn as nn +import torch.nn.functional as F + + +#################### +# RRDBNet Generator +#################### + +class RRDBNet(nn.Module): + def __init__(self, in_nc, out_nc, nf, nb, nr=3, gc=32, upscale=4, norm_type=None, + act_type='leakyrelu', mode='CNA', upsample_mode='upconv', convtype='Conv2D', + finalact=None, gaussian_noise=False, plus=False): + super(RRDBNet, self).__init__() + n_upscale = int(math.log(upscale, 2)) + if upscale == 3: + n_upscale = 1 + + self.resrgan_scale = 0 + if in_nc % 16 == 0: + self.resrgan_scale = 1 + elif in_nc != 4 and in_nc % 4 == 0: + self.resrgan_scale = 2 + + fea_conv = conv_block(in_nc, nf, kernel_size=3, norm_type=None, act_type=None, convtype=convtype) + rb_blocks = [RRDB(nf, nr, kernel_size=3, gc=32, stride=1, bias=1, pad_type='zero', + norm_type=norm_type, act_type=act_type, mode='CNA', convtype=convtype, + gaussian_noise=gaussian_noise, plus=plus) for _ in range(nb)] + LR_conv = conv_block(nf, nf, kernel_size=3, norm_type=norm_type, act_type=None, mode=mode, convtype=convtype) + + if upsample_mode == 'upconv': + upsample_block = upconv_block + elif upsample_mode == 'pixelshuffle': + upsample_block = pixelshuffle_block + else: + raise NotImplementedError(f'upsample mode [{upsample_mode}] is not found') + if upscale == 3: + upsampler = upsample_block(nf, nf, 3, act_type=act_type, convtype=convtype) + else: + upsampler = [upsample_block(nf, nf, act_type=act_type, convtype=convtype) for _ in range(n_upscale)] + HR_conv0 = conv_block(nf, nf, kernel_size=3, norm_type=None, act_type=act_type, convtype=convtype) + HR_conv1 = conv_block(nf, out_nc, kernel_size=3, norm_type=None, act_type=None, convtype=convtype) + + outact = act(finalact) if finalact else None + + self.model = sequential(fea_conv, ShortcutBlock(sequential(*rb_blocks, LR_conv)), + *upsampler, HR_conv0, HR_conv1, outact) + + def forward(self, x, outm=None): + if self.resrgan_scale == 1: + feat = pixel_unshuffle(x, scale=4) + elif self.resrgan_scale == 2: + feat = pixel_unshuffle(x, scale=2) + else: + feat = x + + return self.model(feat) + + +class RRDB(nn.Module): + """ + Residual in Residual Dense Block + (ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks) + """ + + def __init__(self, nf, nr=3, kernel_size=3, gc=32, stride=1, bias=1, pad_type='zero', + norm_type=None, act_type='leakyrelu', mode='CNA', convtype='Conv2D', + spectral_norm=False, gaussian_noise=False, plus=False): + super(RRDB, self).__init__() + # This is for backwards compatibility with existing models + if nr == 3: + self.RDB1 = ResidualDenseBlock_5C(nf, kernel_size, gc, stride, bias, pad_type, + norm_type, act_type, mode, convtype, spectral_norm=spectral_norm, + gaussian_noise=gaussian_noise, plus=plus) + self.RDB2 = ResidualDenseBlock_5C(nf, kernel_size, gc, stride, bias, pad_type, + norm_type, act_type, mode, convtype, spectral_norm=spectral_norm, + gaussian_noise=gaussian_noise, plus=plus) + self.RDB3 = ResidualDenseBlock_5C(nf, kernel_size, gc, stride, bias, pad_type, + norm_type, act_type, mode, convtype, spectral_norm=spectral_norm, + gaussian_noise=gaussian_noise, plus=plus) + else: + RDB_list = [ResidualDenseBlock_5C(nf, kernel_size, gc, stride, bias, pad_type, + norm_type, act_type, mode, convtype, spectral_norm=spectral_norm, + gaussian_noise=gaussian_noise, plus=plus) for _ in range(nr)] + self.RDBs = nn.Sequential(*RDB_list) + + def forward(self, x): + if hasattr(self, 'RDB1'): + out = self.RDB1(x) + out = self.RDB2(out) + out = self.RDB3(out) + else: + out = self.RDBs(x) + return out * 0.2 + x + + +class ResidualDenseBlock_5C(nn.Module): + """ + Residual Dense Block + The core module of paper: (Residual Dense Network for Image Super-Resolution, CVPR 18) + Modified options that can be used: + - "Partial Convolution based Padding" arXiv:1811.11718 + - "Spectral normalization" arXiv:1802.05957 + - "ICASSP 2020 - ESRGAN+ : Further Improving ESRGAN" N. C. + {Rakotonirina} and A. {Rasoanaivo} + """ + + def __init__(self, nf=64, kernel_size=3, gc=32, stride=1, bias=1, pad_type='zero', + norm_type=None, act_type='leakyrelu', mode='CNA', convtype='Conv2D', + spectral_norm=False, gaussian_noise=False, plus=False): + super(ResidualDenseBlock_5C, self).__init__() + + self.noise = GaussianNoise() if gaussian_noise else None + self.conv1x1 = conv1x1(nf, gc) if plus else None + + self.conv1 = conv_block(nf, gc, kernel_size, stride, bias=bias, pad_type=pad_type, + norm_type=norm_type, act_type=act_type, mode=mode, convtype=convtype, + spectral_norm=spectral_norm) + self.conv2 = conv_block(nf+gc, gc, kernel_size, stride, bias=bias, pad_type=pad_type, + norm_type=norm_type, act_type=act_type, mode=mode, convtype=convtype, + spectral_norm=spectral_norm) + self.conv3 = conv_block(nf+2*gc, gc, kernel_size, stride, bias=bias, pad_type=pad_type, + norm_type=norm_type, act_type=act_type, mode=mode, convtype=convtype, + spectral_norm=spectral_norm) + self.conv4 = conv_block(nf+3*gc, gc, kernel_size, stride, bias=bias, pad_type=pad_type, + norm_type=norm_type, act_type=act_type, mode=mode, convtype=convtype, + spectral_norm=spectral_norm) + if mode == 'CNA': + last_act = None + else: + last_act = act_type + self.conv5 = conv_block(nf+4*gc, nf, 3, stride, bias=bias, pad_type=pad_type, + norm_type=norm_type, act_type=last_act, mode=mode, convtype=convtype, + spectral_norm=spectral_norm) + + def forward(self, x): + x1 = self.conv1(x) + x2 = self.conv2(torch.cat((x, x1), 1)) + if self.conv1x1: + x2 = x2 + self.conv1x1(x) + x3 = self.conv3(torch.cat((x, x1, x2), 1)) + x4 = self.conv4(torch.cat((x, x1, x2, x3), 1)) + if self.conv1x1: + x4 = x4 + x2 + x5 = self.conv5(torch.cat((x, x1, x2, x3, x4), 1)) + if self.noise: + return self.noise(x5.mul(0.2) + x) + else: + return x5 * 0.2 + x + + +#################### +# ESRGANplus +#################### + +class GaussianNoise(nn.Module): + def __init__(self, sigma=0.1, is_relative_detach=False): + super().__init__() + self.sigma = sigma + self.is_relative_detach = is_relative_detach + self.noise = torch.tensor(0, dtype=torch.float) + + def forward(self, x): + if self.training and self.sigma != 0: + self.noise = self.noise.to(x.device) + scale = self.sigma * x.detach() if self.is_relative_detach else self.sigma * x + sampled_noise = self.noise.repeat(*x.size()).normal_() * scale + x = x + sampled_noise + return x + +def conv1x1(in_planes, out_planes, stride=1): + return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False) + + +#################### +# SRVGGNetCompact +#################### + +class SRVGGNetCompact(nn.Module): + """A compact VGG-style network structure for super-resolution. + This class is copied from https://github.com/xinntao/Real-ESRGAN + """ + + def __init__(self, num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=16, upscale=4, act_type='prelu'): + super(SRVGGNetCompact, self).__init__() + self.num_in_ch = num_in_ch + self.num_out_ch = num_out_ch + self.num_feat = num_feat + self.num_conv = num_conv + self.upscale = upscale + self.act_type = act_type + + self.body = nn.ModuleList() + # the first conv + self.body.append(nn.Conv2d(num_in_ch, num_feat, 3, 1, 1)) + # the first activation + if act_type == 'relu': + activation = nn.ReLU(inplace=True) + elif act_type == 'prelu': + activation = nn.PReLU(num_parameters=num_feat) + elif act_type == 'leakyrelu': + activation = nn.LeakyReLU(negative_slope=0.1, inplace=True) + self.body.append(activation) + + # the body structure + for _ in range(num_conv): + self.body.append(nn.Conv2d(num_feat, num_feat, 3, 1, 1)) + # activation + if act_type == 'relu': + activation = nn.ReLU(inplace=True) + elif act_type == 'prelu': + activation = nn.PReLU(num_parameters=num_feat) + elif act_type == 'leakyrelu': + activation = nn.LeakyReLU(negative_slope=0.1, inplace=True) + self.body.append(activation) + + # the last conv + self.body.append(nn.Conv2d(num_feat, num_out_ch * upscale * upscale, 3, 1, 1)) + # upsample + self.upsampler = nn.PixelShuffle(upscale) + + def forward(self, x): + out = x + for i in range(0, len(self.body)): + out = self.body[i](out) + + out = self.upsampler(out) + # add the nearest upsampled image, so that the network learns the residual + base = F.interpolate(x, scale_factor=self.upscale, mode='nearest') + out += base + return out + + +#################### +# Upsampler +#################### + +class Upsample(nn.Module): + r"""Upsamples a given multi-channel 1D (temporal), 2D (spatial) or 3D (volumetric) data. + The input data is assumed to be of the form + `minibatch x channels x [optional depth] x [optional height] x width`. + """ + + def __init__(self, size=None, scale_factor=None, mode="nearest", align_corners=None): + super(Upsample, self).__init__() + if isinstance(scale_factor, tuple): + self.scale_factor = tuple(float(factor) for factor in scale_factor) + else: + self.scale_factor = float(scale_factor) if scale_factor else None + self.mode = mode + self.size = size + self.align_corners = align_corners + + def forward(self, x): + return nn.functional.interpolate(x, size=self.size, scale_factor=self.scale_factor, mode=self.mode, align_corners=self.align_corners) + + def extra_repr(self): + if self.scale_factor is not None: + info = f'scale_factor={self.scale_factor}' + else: + info = f'size={self.size}' + info += f', mode={self.mode}' + return info + + +def pixel_unshuffle(x, scale): + """ Pixel unshuffle. + Args: + x (Tensor): Input feature with shape (b, c, hh, hw). + scale (int): Downsample ratio. + Returns: + Tensor: the pixel unshuffled feature. + """ + b, c, hh, hw = x.size() + out_channel = c * (scale**2) + assert hh % scale == 0 and hw % scale == 0 + h = hh // scale + w = hw // scale + x_view = x.view(b, c, h, scale, w, scale) + return x_view.permute(0, 1, 3, 5, 2, 4).reshape(b, out_channel, h, w) + + +def pixelshuffle_block(in_nc, out_nc, upscale_factor=2, kernel_size=3, stride=1, bias=True, + pad_type='zero', norm_type=None, act_type='relu', convtype='Conv2D'): + """ + Pixel shuffle layer + (Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional + Neural Network, CVPR17) + """ + conv = conv_block(in_nc, out_nc * (upscale_factor ** 2), kernel_size, stride, bias=bias, + pad_type=pad_type, norm_type=None, act_type=None, convtype=convtype) + pixel_shuffle = nn.PixelShuffle(upscale_factor) + + n = norm(norm_type, out_nc) if norm_type else None + a = act(act_type) if act_type else None + return sequential(conv, pixel_shuffle, n, a) + + +def upconv_block(in_nc, out_nc, upscale_factor=2, kernel_size=3, stride=1, bias=True, + pad_type='zero', norm_type=None, act_type='relu', mode='nearest', convtype='Conv2D'): + """ Upconv layer """ + upscale_factor = (1, upscale_factor, upscale_factor) if convtype == 'Conv3D' else upscale_factor + upsample = Upsample(scale_factor=upscale_factor, mode=mode) + conv = conv_block(in_nc, out_nc, kernel_size, stride, bias=bias, + pad_type=pad_type, norm_type=norm_type, act_type=act_type, convtype=convtype) + return sequential(upsample, conv) + + + + + + + + +#################### +# Basic blocks +#################### + + +def make_layer(basic_block, num_basic_block, **kwarg): + """Make layers by stacking the same blocks. + Args: + basic_block (nn.module): nn.module class for basic block. (block) + num_basic_block (int): number of blocks. (n_layers) + Returns: + nn.Sequential: Stacked blocks in nn.Sequential. + """ + layers = [] + for _ in range(num_basic_block): + layers.append(basic_block(**kwarg)) + return nn.Sequential(*layers) + + +def act(act_type, inplace=True, neg_slope=0.2, n_prelu=1, beta=1.0): + """ activation helper """ + act_type = act_type.lower() + if act_type == 'relu': + layer = nn.ReLU(inplace) + elif act_type in ('leakyrelu', 'lrelu'): + layer = nn.LeakyReLU(neg_slope, inplace) + elif act_type == 'prelu': + layer = nn.PReLU(num_parameters=n_prelu, init=neg_slope) + elif act_type == 'tanh': # [-1, 1] range output + layer = nn.Tanh() + elif act_type == 'sigmoid': # [0, 1] range output + layer = nn.Sigmoid() + else: + raise NotImplementedError(f'activation layer [{act_type}] is not found') + return layer + + +class Identity(nn.Module): + def __init__(self, *kwargs): + super(Identity, self).__init__() + + def forward(self, x, *kwargs): + return x + + +def norm(norm_type, nc): + """ Return a normalization layer """ + norm_type = norm_type.lower() + if norm_type == 'batch': + layer = nn.BatchNorm2d(nc, affine=True) + elif norm_type == 'instance': + layer = nn.InstanceNorm2d(nc, affine=False) + elif norm_type == 'none': + def norm_layer(x): return Identity() + else: + raise NotImplementedError(f'normalization layer [{norm_type}] is not found') + return layer + + +def pad(pad_type, padding): + """ padding layer helper """ + pad_type = pad_type.lower() + if padding == 0: + return None + if pad_type == 'reflect': + layer = nn.ReflectionPad2d(padding) + elif pad_type == 'replicate': + layer = nn.ReplicationPad2d(padding) + elif pad_type == 'zero': + layer = nn.ZeroPad2d(padding) + else: + raise NotImplementedError(f'padding layer [{pad_type}] is not implemented') + return layer + + +def get_valid_padding(kernel_size, dilation): + kernel_size = kernel_size + (kernel_size - 1) * (dilation - 1) + padding = (kernel_size - 1) // 2 + return padding + + +class ShortcutBlock(nn.Module): + """ Elementwise sum the output of a submodule to its input """ + def __init__(self, submodule): + super(ShortcutBlock, self).__init__() + self.sub = submodule + + def forward(self, x): + output = x + self.sub(x) + return output + + def __repr__(self): + return 'Identity + \n|' + self.sub.__repr__().replace('\n', '\n|') + + +def sequential(*args): + """ Flatten Sequential. It unwraps nn.Sequential. """ + if len(args) == 1: + if isinstance(args[0], OrderedDict): + raise NotImplementedError('sequential does not support OrderedDict input.') + return args[0] # No sequential is needed. + modules = [] + for module in args: + if isinstance(module, nn.Sequential): + for submodule in module.children(): + modules.append(submodule) + elif isinstance(module, nn.Module): + modules.append(module) + return nn.Sequential(*modules) + + +def conv_block(in_nc, out_nc, kernel_size, stride=1, dilation=1, groups=1, bias=True, + pad_type='zero', norm_type=None, act_type='relu', mode='CNA', convtype='Conv2D', + spectral_norm=False): + """ Conv layer with padding, normalization, activation """ + assert mode in ['CNA', 'NAC', 'CNAC'], f'Wrong conv mode [{mode}]' + padding = get_valid_padding(kernel_size, dilation) + p = pad(pad_type, padding) if pad_type and pad_type != 'zero' else None + padding = padding if pad_type == 'zero' else 0 + + if convtype=='PartialConv2D': + from torchvision.ops import PartialConv2d # this is definitely not going to work, but PartialConv2d doesn't work anyway and this shuts up static analyzer + c = PartialConv2d(in_nc, out_nc, kernel_size=kernel_size, stride=stride, padding=padding, + dilation=dilation, bias=bias, groups=groups) + elif convtype=='DeformConv2D': + from torchvision.ops import DeformConv2d # not tested + c = DeformConv2d(in_nc, out_nc, kernel_size=kernel_size, stride=stride, padding=padding, + dilation=dilation, bias=bias, groups=groups) + elif convtype=='Conv3D': + c = nn.Conv3d(in_nc, out_nc, kernel_size=kernel_size, stride=stride, padding=padding, + dilation=dilation, bias=bias, groups=groups) + else: + c = nn.Conv2d(in_nc, out_nc, kernel_size=kernel_size, stride=stride, padding=padding, + dilation=dilation, bias=bias, groups=groups) + + if spectral_norm: + c = nn.utils.spectral_norm(c) + + a = act(act_type) if act_type else None + if 'CNA' in mode: + n = norm(norm_type, out_nc) if norm_type else None + return sequential(p, c, n, a) + elif mode == 'NAC': + if norm_type is None and act_type is not None: + a = act(act_type, inplace=False) + n = norm(norm_type, in_nc) if norm_type else None + return sequential(n, a, p, c) diff --git a/modules/extensions.py b/modules/extensions.py new file mode 100644 index 0000000000000000000000000000000000000000..8f4755b8fe161698d938012fd7d693e7325f8101 --- /dev/null +++ b/modules/extensions.py @@ -0,0 +1,152 @@ +import os +import sys +import threading +import traceback + +import git + +from modules import shared +from modules.paths_internal import extensions_dir, extensions_builtin_dir, script_path # noqa: F401 + +extensions = [] + +if not os.path.exists(extensions_dir): + os.makedirs(extensions_dir) + + +def active(): + if shared.opts.disable_all_extensions == "all": + return [] + elif shared.opts.disable_all_extensions == "extra": + return [x for x in extensions if x.enabled and x.is_builtin] + else: + return [x for x in extensions if x.enabled] + + +class Extension: + lock = threading.Lock() + + def __init__(self, name, path, enabled=True, is_builtin=False): + self.name = name + self.path = path + self.enabled = enabled + self.status = '' + self.can_update = False + self.is_builtin = is_builtin + self.commit_hash = '' + self.commit_date = None + self.version = '' + self.branch = None + self.remote = None + self.have_info_from_repo = False + + def read_info_from_repo(self): + if self.is_builtin or self.have_info_from_repo: + return + + with self.lock: + if self.have_info_from_repo: + return + + self.do_read_info_from_repo() + + def do_read_info_from_repo(self): + repo = None + try: + if os.path.exists(os.path.join(self.path, ".git")): + repo = git.Repo(self.path) + except Exception: + print(f"Error reading github repository info from {self.path}:", file=sys.stderr) + print(traceback.format_exc(), file=sys.stderr) + + if repo is None or repo.bare: + self.remote = None + else: + try: + self.status = 'unknown' + self.remote = next(repo.remote().urls, None) + commit = repo.head.commit + self.commit_date = commit.committed_date + if repo.active_branch: + self.branch = repo.active_branch.name + self.commit_hash = commit.hexsha + self.version = self.commit_hash[:8] + + except Exception as ex: + print(f"Failed reading extension data from Git repository ({self.name}): {ex}", file=sys.stderr) + self.remote = None + + self.have_info_from_repo = True + + def list_files(self, subdir, extension): + from modules import scripts + + dirpath = os.path.join(self.path, subdir) + if not os.path.isdir(dirpath): + return [] + + res = [] + for filename in sorted(os.listdir(dirpath)): + res.append(scripts.ScriptFile(self.path, filename, os.path.join(dirpath, filename))) + + res = [x for x in res if os.path.splitext(x.path)[1].lower() == extension and os.path.isfile(x.path)] + + return res + + def check_updates(self): + repo = git.Repo(self.path) + for fetch in repo.remote().fetch(dry_run=True): + if fetch.flags != fetch.HEAD_UPTODATE: + self.can_update = True + self.status = "new commits" + return + + try: + origin = repo.rev_parse('origin') + if repo.head.commit != origin: + self.can_update = True + self.status = "behind HEAD" + return + except Exception: + self.can_update = False + self.status = "unknown (remote error)" + return + + self.can_update = False + self.status = "latest" + + def fetch_and_reset_hard(self, commit='origin'): + repo = git.Repo(self.path) + # Fix: `error: Your local changes to the following files would be overwritten by merge`, + # because WSL2 Docker set 755 file permissions instead of 644, this results to the error. + repo.git.fetch(all=True) + repo.git.reset(commit, hard=True) + self.have_info_from_repo = False + + +def list_extensions(): + extensions.clear() + + if not os.path.isdir(extensions_dir): + return + + if shared.opts.disable_all_extensions == "all": + print("*** \"Disable all extensions\" option was set, will not load any extensions ***") + elif shared.opts.disable_all_extensions == "extra": + print("*** \"Disable all extensions\" option was set, will only load built-in extensions ***") + + extension_paths = [] + for dirname in [extensions_dir, extensions_builtin_dir]: + if not os.path.isdir(dirname): + return + + for extension_dirname in sorted(os.listdir(dirname)): + path = os.path.join(dirname, extension_dirname) + if not os.path.isdir(path): + continue + + extension_paths.append((extension_dirname, path, dirname == extensions_builtin_dir)) + + for dirname, path, is_builtin in extension_paths: + extension = Extension(name=dirname, path=path, enabled=dirname not in shared.opts.disabled_extensions, is_builtin=is_builtin) + extensions.append(extension) diff --git a/modules/extra_networks.py b/modules/extra_networks.py new file mode 100644 index 0000000000000000000000000000000000000000..e8e8ca39b7490588a99793667d5fab7718ff958c --- /dev/null +++ b/modules/extra_networks.py @@ -0,0 +1,161 @@ +import re +from collections import defaultdict + +from modules import errors + +extra_network_registry = {} + + +def initialize(): + extra_network_registry.clear() + + +def register_extra_network(extra_network): + extra_network_registry[extra_network.name] = extra_network + + +def register_default_extra_networks(): + from modules.extra_networks_hypernet import ExtraNetworkHypernet + register_extra_network(ExtraNetworkHypernet()) + + +class ExtraNetworkParams: + def __init__(self, items=None): + self.items = items or [] + self.positional = [] + self.named = {} + + for item in self.items: + parts = item.split('=', 2) if isinstance(item, str) else [item] + if len(parts) == 2: + self.named[parts[0]] = parts[1] + else: + self.positional.append(item) + + +class ExtraNetwork: + def __init__(self, name): + self.name = name + + def activate(self, p, params_list): + """ + Called by processing on every run. Whatever the extra network is meant to do should be activated here. + Passes arguments related to this extra network in params_list. + User passes arguments by specifying this in his prompt: + + + + Where name matches the name of this ExtraNetwork object, and arg1:arg2:arg3 are any natural number of text arguments + separated by colon. + + Even if the user does not mention this ExtraNetwork in his prompt, the call will stil be made, with empty params_list - + in this case, all effects of this extra networks should be disabled. + + Can be called multiple times before deactivate() - each new call should override the previous call completely. + + For example, if this ExtraNetwork's name is 'hypernet' and user's prompt is: + + > "1girl, " + + params_list will be: + + [ + ExtraNetworkParams(items=["agm", "1.1"]), + ExtraNetworkParams(items=["ray"]) + ] + + """ + raise NotImplementedError + + def deactivate(self, p): + """ + Called at the end of processing for housekeeping. No need to do anything here. + """ + + raise NotImplementedError + + +def activate(p, extra_network_data): + """call activate for extra networks in extra_network_data in specified order, then call + activate for all remaining registered networks with an empty argument list""" + + for extra_network_name, extra_network_args in extra_network_data.items(): + extra_network = extra_network_registry.get(extra_network_name, None) + if extra_network is None: + print(f"Skipping unknown extra network: {extra_network_name}") + continue + + try: + extra_network.activate(p, extra_network_args) + except Exception as e: + errors.display(e, f"activating extra network {extra_network_name} with arguments {extra_network_args}") + + for extra_network_name, extra_network in extra_network_registry.items(): + args = extra_network_data.get(extra_network_name, None) + if args is not None: + continue + + try: + extra_network.activate(p, []) + except Exception as e: + errors.display(e, f"activating extra network {extra_network_name}") + + +def deactivate(p, extra_network_data): + """call deactivate for extra networks in extra_network_data in specified order, then call + deactivate for all remaining registered networks""" + + for extra_network_name in extra_network_data: + extra_network = extra_network_registry.get(extra_network_name, None) + if extra_network is None: + continue + + try: + extra_network.deactivate(p) + except Exception as e: + errors.display(e, f"deactivating extra network {extra_network_name}") + + for extra_network_name, extra_network in extra_network_registry.items(): + args = extra_network_data.get(extra_network_name, None) + if args is not None: + continue + + try: + extra_network.deactivate(p) + except Exception as e: + errors.display(e, f"deactivating unmentioned extra network {extra_network_name}") + + +re_extra_net = re.compile(r"<(\w+):([^>]+)>") + + +def parse_prompt(prompt): + res = defaultdict(list) + + def found(m): + name = m.group(1) + args = m.group(2) + + res[name].append(ExtraNetworkParams(items=args.split(":"))) + + return "" + + prompt = re.sub(re_extra_net, found, prompt) + + return prompt, res + + +def parse_prompts(prompts): + res = [] + extra_data = None + + for prompt in prompts: + updated_prompt, parsed_extra_data = parse_prompt(prompt) + + if extra_data is None: + extra_data = parsed_extra_data + + res.append(updated_prompt) + + return res, extra_data + diff --git a/modules/extra_networks_hypernet.py b/modules/extra_networks_hypernet.py new file mode 100644 index 0000000000000000000000000000000000000000..dce11b68a31cdbfaf08e24dcbb93c59a2e9df88b --- /dev/null +++ b/modules/extra_networks_hypernet.py @@ -0,0 +1,28 @@ +from modules import extra_networks, shared +from modules.hypernetworks import hypernetwork + + +class ExtraNetworkHypernet(extra_networks.ExtraNetwork): + def __init__(self): + super().__init__('hypernet') + + def activate(self, p, params_list): + additional = shared.opts.sd_hypernetwork + + if additional != "None" and additional in shared.hypernetworks and len([x for x in params_list if x.items[0] == additional]) == 0: + hypernet_prompt_text = f"" + p.all_prompts = [f"{prompt}{hypernet_prompt_text}" for prompt in p.all_prompts] + params_list.append(extra_networks.ExtraNetworkParams(items=[additional, shared.opts.extra_networks_default_multiplier])) + + names = [] + multipliers = [] + for params in params_list: + assert len(params.items) > 0 + + names.append(params.items[0]) + multipliers.append(float(params.items[1]) if len(params.items) > 1 else 1.0) + + hypernetwork.load_hypernetworks(names, multipliers) + + def deactivate(self, p): + pass diff --git a/modules/extras.py b/modules/extras.py new file mode 100644 index 0000000000000000000000000000000000000000..268befb7b6e2212b699fa6630e21f7fe9e8763c5 --- /dev/null +++ b/modules/extras.py @@ -0,0 +1,304 @@ +import os +import re +import shutil +import json + + +import torch +import tqdm + +from modules import shared, images, sd_models, sd_vae, sd_models_config +from modules.ui_common import plaintext_to_html +import gradio as gr +import safetensors.torch + + +def run_pnginfo(image): + if image is None: + return '', '', '' + + geninfo, items = images.read_info_from_image(image) + items = {**{'parameters': geninfo}, **items} + + info = '' + for key, text in items.items(): + info += f""" +
+

{plaintext_to_html(str(key))}

+

{plaintext_to_html(str(text))}

+
+""".strip()+"\n" + + if len(info) == 0: + message = "Nothing found in the image." + info = f"

{message}

" + + return '', geninfo, info + + +def create_config(ckpt_result, config_source, a, b, c): + def config(x): + res = sd_models_config.find_checkpoint_config_near_filename(x) if x else None + return res if res != shared.sd_default_config else None + + if config_source == 0: + cfg = config(a) or config(b) or config(c) + elif config_source == 1: + cfg = config(b) + elif config_source == 2: + cfg = config(c) + else: + cfg = None + + if cfg is None: + return + + filename, _ = os.path.splitext(ckpt_result) + checkpoint_filename = filename + ".yaml" + + print("Copying config:") + print(" from:", cfg) + print(" to:", checkpoint_filename) + shutil.copyfile(cfg, checkpoint_filename) + + +checkpoint_dict_skip_on_merge = ["cond_stage_model.transformer.text_model.embeddings.position_ids"] + + +def to_half(tensor, enable): + if enable and tensor.dtype == torch.float: + return tensor.half() + + return tensor + + +def run_modelmerger(id_task, primary_model_name, secondary_model_name, tertiary_model_name, interp_method, multiplier, save_as_half, custom_name, checkpoint_format, config_source, bake_in_vae, discard_weights, save_metadata): + shared.state.begin() + shared.state.job = 'model-merge' + + def fail(message): + shared.state.textinfo = message + shared.state.end() + return [*[gr.update() for _ in range(4)], message] + + def weighted_sum(theta0, theta1, alpha): + return ((1 - alpha) * theta0) + (alpha * theta1) + + def get_difference(theta1, theta2): + return theta1 - theta2 + + def add_difference(theta0, theta1_2_diff, alpha): + return theta0 + (alpha * theta1_2_diff) + + def filename_weighted_sum(): + a = primary_model_info.model_name + b = secondary_model_info.model_name + Ma = round(1 - multiplier, 2) + Mb = round(multiplier, 2) + + return f"{Ma}({a}) + {Mb}({b})" + + def filename_add_difference(): + a = primary_model_info.model_name + b = secondary_model_info.model_name + c = tertiary_model_info.model_name + M = round(multiplier, 2) + + return f"{a} + {M}({b} - {c})" + + def filename_nothing(): + return primary_model_info.model_name + + theta_funcs = { + "Weighted sum": (filename_weighted_sum, None, weighted_sum), + "Add difference": (filename_add_difference, get_difference, add_difference), + "No interpolation": (filename_nothing, None, None), + } + filename_generator, theta_func1, theta_func2 = theta_funcs[interp_method] + shared.state.job_count = (1 if theta_func1 else 0) + (1 if theta_func2 else 0) + + if not primary_model_name: + return fail("Failed: Merging requires a primary model.") + + primary_model_info = sd_models.checkpoints_list[primary_model_name] + + if theta_func2 and not secondary_model_name: + return fail("Failed: Merging requires a secondary model.") + + secondary_model_info = sd_models.checkpoints_list[secondary_model_name] if theta_func2 else None + + if theta_func1 and not tertiary_model_name: + return fail(f"Failed: Interpolation method ({interp_method}) requires a tertiary model.") + + tertiary_model_info = sd_models.checkpoints_list[tertiary_model_name] if theta_func1 else None + + result_is_inpainting_model = False + result_is_instruct_pix2pix_model = False + + if theta_func2: + shared.state.textinfo = "Loading B" + print(f"Loading {secondary_model_info.filename}...") + theta_1 = sd_models.read_state_dict(secondary_model_info.filename, map_location='cpu') + else: + theta_1 = None + + if theta_func1: + shared.state.textinfo = "Loading C" + print(f"Loading {tertiary_model_info.filename}...") + theta_2 = sd_models.read_state_dict(tertiary_model_info.filename, map_location='cpu') + + shared.state.textinfo = 'Merging B and C' + shared.state.sampling_steps = len(theta_1.keys()) + for key in tqdm.tqdm(theta_1.keys()): + if key in checkpoint_dict_skip_on_merge: + continue + + if 'model' in key: + if key in theta_2: + t2 = theta_2.get(key, torch.zeros_like(theta_1[key])) + theta_1[key] = theta_func1(theta_1[key], t2) + else: + theta_1[key] = torch.zeros_like(theta_1[key]) + + shared.state.sampling_step += 1 + del theta_2 + + shared.state.nextjob() + + shared.state.textinfo = f"Loading {primary_model_info.filename}..." + print(f"Loading {primary_model_info.filename}...") + theta_0 = sd_models.read_state_dict(primary_model_info.filename, map_location='cpu') + + print("Merging...") + shared.state.textinfo = 'Merging A and B' + shared.state.sampling_steps = len(theta_0.keys()) + for key in tqdm.tqdm(theta_0.keys()): + if theta_1 and 'model' in key and key in theta_1: + + if key in checkpoint_dict_skip_on_merge: + continue + + a = theta_0[key] + b = theta_1[key] + + # this enables merging an inpainting model (A) with another one (B); + # where normal model would have 4 channels, for latenst space, inpainting model would + # have another 4 channels for unmasked picture's latent space, plus one channel for mask, for a total of 9 + if a.shape != b.shape and a.shape[0:1] + a.shape[2:] == b.shape[0:1] + b.shape[2:]: + if a.shape[1] == 4 and b.shape[1] == 9: + raise RuntimeError("When merging inpainting model with a normal one, A must be the inpainting model.") + if a.shape[1] == 4 and b.shape[1] == 8: + raise RuntimeError("When merging instruct-pix2pix model with a normal one, A must be the instruct-pix2pix model.") + + if a.shape[1] == 8 and b.shape[1] == 4:#If we have an Instruct-Pix2Pix model... + theta_0[key][:, 0:4, :, :] = theta_func2(a[:, 0:4, :, :], b, multiplier)#Merge only the vectors the models have in common. Otherwise we get an error due to dimension mismatch. + result_is_instruct_pix2pix_model = True + else: + assert a.shape[1] == 9 and b.shape[1] == 4, f"Bad dimensions for merged layer {key}: A={a.shape}, B={b.shape}" + theta_0[key][:, 0:4, :, :] = theta_func2(a[:, 0:4, :, :], b, multiplier) + result_is_inpainting_model = True + else: + theta_0[key] = theta_func2(a, b, multiplier) + + theta_0[key] = to_half(theta_0[key], save_as_half) + + shared.state.sampling_step += 1 + + del theta_1 + + bake_in_vae_filename = sd_vae.vae_dict.get(bake_in_vae, None) + if bake_in_vae_filename is not None: + print(f"Baking in VAE from {bake_in_vae_filename}") + shared.state.textinfo = 'Baking in VAE' + vae_dict = sd_vae.load_vae_dict(bake_in_vae_filename, map_location='cpu') + + for key in vae_dict.keys(): + theta_0_key = 'first_stage_model.' + key + if theta_0_key in theta_0: + theta_0[theta_0_key] = to_half(vae_dict[key], save_as_half) + + del vae_dict + + if save_as_half and not theta_func2: + for key in theta_0.keys(): + theta_0[key] = to_half(theta_0[key], save_as_half) + + if discard_weights: + regex = re.compile(discard_weights) + for key in list(theta_0): + if re.search(regex, key): + theta_0.pop(key, None) + + ckpt_dir = shared.cmd_opts.ckpt_dir or sd_models.model_path + + filename = filename_generator() if custom_name == '' else custom_name + filename += ".inpainting" if result_is_inpainting_model else "" + filename += ".instruct-pix2pix" if result_is_instruct_pix2pix_model else "" + filename += "." + checkpoint_format + + output_modelname = os.path.join(ckpt_dir, filename) + + shared.state.nextjob() + shared.state.textinfo = "Saving" + print(f"Saving to {output_modelname}...") + + metadata = None + + if save_metadata: + metadata = {"format": "pt"} + + merge_recipe = { + "type": "webui", # indicate this model was merged with webui's built-in merger + "primary_model_hash": primary_model_info.sha256, + "secondary_model_hash": secondary_model_info.sha256 if secondary_model_info else None, + "tertiary_model_hash": tertiary_model_info.sha256 if tertiary_model_info else None, + "interp_method": interp_method, + "multiplier": multiplier, + "save_as_half": save_as_half, + "custom_name": custom_name, + "config_source": config_source, + "bake_in_vae": bake_in_vae, + "discard_weights": discard_weights, + "is_inpainting": result_is_inpainting_model, + "is_instruct_pix2pix": result_is_instruct_pix2pix_model + } + metadata["sd_merge_recipe"] = json.dumps(merge_recipe) + + sd_merge_models = {} + + def add_model_metadata(checkpoint_info): + checkpoint_info.calculate_shorthash() + sd_merge_models[checkpoint_info.sha256] = { + "name": checkpoint_info.name, + "legacy_hash": checkpoint_info.hash, + "sd_merge_recipe": checkpoint_info.metadata.get("sd_merge_recipe", None) + } + + sd_merge_models.update(checkpoint_info.metadata.get("sd_merge_models", {})) + + add_model_metadata(primary_model_info) + if secondary_model_info: + add_model_metadata(secondary_model_info) + if tertiary_model_info: + add_model_metadata(tertiary_model_info) + + metadata["sd_merge_models"] = json.dumps(sd_merge_models) + + _, extension = os.path.splitext(output_modelname) + if extension.lower() == ".safetensors": + safetensors.torch.save_file(theta_0, output_modelname, metadata=metadata) + else: + torch.save(theta_0, output_modelname) + + sd_models.list_models() + created_model = next((ckpt for ckpt in sd_models.checkpoints_list.values() if ckpt.name == filename), None) + if created_model: + created_model.calculate_shorthash() + + create_config(output_modelname, config_source, primary_model_info, secondary_model_info, tertiary_model_info) + + print(f"Checkpoint saved to {output_modelname}.") + shared.state.textinfo = "Checkpoint saved" + shared.state.end() + + return [*[gr.Dropdown.update(choices=sd_models.checkpoint_tiles()) for _ in range(4)], "Checkpoint saved to " + output_modelname] diff --git a/modules/face_restoration.py b/modules/face_restoration.py new file mode 100644 index 0000000000000000000000000000000000000000..2c86c6ccce338a1411f4367a0bc6e4046ad67cae --- /dev/null +++ b/modules/face_restoration.py @@ -0,0 +1,19 @@ +from modules import shared + + +class FaceRestoration: + def name(self): + return "None" + + def restore(self, np_image): + return np_image + + +def restore_faces(np_image): + face_restorers = [x for x in shared.face_restorers if x.name() == shared.opts.face_restoration_model or shared.opts.face_restoration_model is None] + if len(face_restorers) == 0: + return np_image + + face_restorer = face_restorers[0] + + return face_restorer.restore(np_image) diff --git a/modules/generation_parameters_copypaste.py b/modules/generation_parameters_copypaste.py new file mode 100644 index 0000000000000000000000000000000000000000..26cdb61da14f59334ceadad1d8656d773515d710 --- /dev/null +++ b/modules/generation_parameters_copypaste.py @@ -0,0 +1,442 @@ +import base64 +import io +import json +import os +import re + +import gradio as gr +from modules.paths import data_path +from modules import shared, ui_tempdir, script_callbacks +from PIL import Image + +re_param_code = r'\s*([\w ]+):\s*("(?:\\"[^,]|\\"|\\|[^\"])+"|[^,]*)(?:,|$)' +re_param = re.compile(re_param_code) +re_imagesize = re.compile(r"^(\d+)x(\d+)$") +re_hypernet_hash = re.compile("\(([0-9a-f]+)\)$") +type_of_gr_update = type(gr.update()) + +paste_fields = {} +registered_param_bindings = [] + + +class ParamBinding: + def __init__(self, paste_button, tabname, source_text_component=None, source_image_component=None, source_tabname=None, override_settings_component=None, paste_field_names=None): + self.paste_button = paste_button + self.tabname = tabname + self.source_text_component = source_text_component + self.source_image_component = source_image_component + self.source_tabname = source_tabname + self.override_settings_component = override_settings_component + self.paste_field_names = paste_field_names or [] + + +def reset(): + paste_fields.clear() + + +def quote(text): + if ',' not in str(text) and '\n' not in str(text) and ':' not in str(text): + return text + + return json.dumps(text, ensure_ascii=False) + + +def unquote(text): + if len(text) == 0 or text[0] != '"' or text[-1] != '"': + return text + + try: + return json.loads(text) + except Exception: + return text + + +def image_from_url_text(filedata): + if filedata is None: + return None + + if type(filedata) == list and len(filedata) > 0 and type(filedata[0]) == dict and filedata[0].get("is_file", False): + filedata = filedata[0] + + if type(filedata) == dict and filedata.get("is_file", False): + filename = filedata["name"] + is_in_right_dir = ui_tempdir.check_tmp_file(shared.demo, filename) + assert is_in_right_dir, 'trying to open image file outside of allowed directories' + + filename = filename.rsplit('?', 1)[0] + return Image.open(filename) + + if type(filedata) == list: + if len(filedata) == 0: + return None + + filedata = filedata[0] + + if filedata.startswith("data:image/png;base64,"): + filedata = filedata[len("data:image/png;base64,"):] + + filedata = base64.decodebytes(filedata.encode('utf-8')) + image = Image.open(io.BytesIO(filedata)) + return image + + +def add_paste_fields(tabname, init_img, fields, override_settings_component=None): + paste_fields[tabname] = {"init_img": init_img, "fields": fields, "override_settings_component": override_settings_component} + + # backwards compatibility for existing extensions + import modules.ui + if tabname == 'txt2img': + modules.ui.txt2img_paste_fields = fields + elif tabname == 'img2img': + modules.ui.img2img_paste_fields = fields + + +def create_buttons(tabs_list): + buttons = {} + for tab in tabs_list: + buttons[tab] = gr.Button(f"Send to {tab}", elem_id=f"{tab}_tab") + return buttons + + +def bind_buttons(buttons, send_image, send_generate_info): + """old function for backwards compatibility; do not use this, use register_paste_params_button""" + for tabname, button in buttons.items(): + source_text_component = send_generate_info if isinstance(send_generate_info, gr.components.Component) else None + source_tabname = send_generate_info if isinstance(send_generate_info, str) else None + + register_paste_params_button(ParamBinding(paste_button=button, tabname=tabname, source_text_component=source_text_component, source_image_component=send_image, source_tabname=source_tabname)) + + +def register_paste_params_button(binding: ParamBinding): + registered_param_bindings.append(binding) + + +def connect_paste_params_buttons(): + binding: ParamBinding + for binding in registered_param_bindings: + destination_image_component = paste_fields[binding.tabname]["init_img"] + fields = paste_fields[binding.tabname]["fields"] + override_settings_component = binding.override_settings_component or paste_fields[binding.tabname]["override_settings_component"] + + destination_width_component = next(iter([field for field, name in fields if name == "Size-1"] if fields else []), None) + destination_height_component = next(iter([field for field, name in fields if name == "Size-2"] if fields else []), None) + + if binding.source_image_component and destination_image_component: + if isinstance(binding.source_image_component, gr.Gallery): + func = send_image_and_dimensions if destination_width_component else image_from_url_text + jsfunc = "extract_image_from_gallery" + else: + func = send_image_and_dimensions if destination_width_component else lambda x: x + jsfunc = None + + binding.paste_button.click( + fn=func, + _js=jsfunc, + inputs=[binding.source_image_component], + outputs=[destination_image_component, destination_width_component, destination_height_component] if destination_width_component else [destination_image_component], + show_progress=False, + ) + + if binding.source_text_component is not None and fields is not None: + connect_paste(binding.paste_button, fields, binding.source_text_component, override_settings_component, binding.tabname) + + if binding.source_tabname is not None and fields is not None: + paste_field_names = ['Prompt', 'Negative prompt', 'Steps', 'Face restoration'] + (["Seed"] if shared.opts.send_seed else []) + binding.paste_field_names + binding.paste_button.click( + fn=lambda *x: x, + inputs=[field for field, name in paste_fields[binding.source_tabname]["fields"] if name in paste_field_names], + outputs=[field for field, name in fields if name in paste_field_names], + show_progress=False, + ) + + binding.paste_button.click( + fn=None, + _js=f"switch_to_{binding.tabname}", + inputs=None, + outputs=None, + show_progress=False, + ) + + +def send_image_and_dimensions(x): + if isinstance(x, Image.Image): + img = x + else: + img = image_from_url_text(x) + + if shared.opts.send_size and isinstance(img, Image.Image): + w = img.width + h = img.height + else: + w = gr.update() + h = gr.update() + + return img, w, h + + + +def find_hypernetwork_key(hypernet_name, hypernet_hash=None): + """Determines the config parameter name to use for the hypernet based on the parameters in the infotext. + + Example: an infotext provides "Hypernet: ke-ta" and "Hypernet hash: 1234abcd". For the "Hypernet" config + parameter this means there should be an entry that looks like "ke-ta-10000(1234abcd)" to set it to. + + If the infotext has no hash, then a hypernet with the same name will be selected instead. + """ + hypernet_name = hypernet_name.lower() + if hypernet_hash is not None: + # Try to match the hash in the name + for hypernet_key in shared.hypernetworks.keys(): + result = re_hypernet_hash.search(hypernet_key) + if result is not None and result[1] == hypernet_hash: + return hypernet_key + else: + # Fall back to a hypernet with the same name + for hypernet_key in shared.hypernetworks.keys(): + if hypernet_key.lower().startswith(hypernet_name): + return hypernet_key + + return None + + +def restore_old_hires_fix_params(res): + """for infotexts that specify old First pass size parameter, convert it into + width, height, and hr scale""" + + firstpass_width = res.get('First pass size-1', None) + firstpass_height = res.get('First pass size-2', None) + + if shared.opts.use_old_hires_fix_width_height: + hires_width = int(res.get("Hires resize-1", 0)) + hires_height = int(res.get("Hires resize-2", 0)) + + if hires_width and hires_height: + res['Size-1'] = hires_width + res['Size-2'] = hires_height + return + + if firstpass_width is None or firstpass_height is None: + return + + firstpass_width, firstpass_height = int(firstpass_width), int(firstpass_height) + width = int(res.get("Size-1", 512)) + height = int(res.get("Size-2", 512)) + + if firstpass_width == 0 or firstpass_height == 0: + from modules import processing + firstpass_width, firstpass_height = processing.old_hires_fix_first_pass_dimensions(width, height) + + res['Size-1'] = firstpass_width + res['Size-2'] = firstpass_height + res['Hires resize-1'] = width + res['Hires resize-2'] = height + + +def parse_generation_parameters(x: str): + """parses generation parameters string, the one you see in text field under the picture in UI: +``` +girl with an artist's beret, determined, blue eyes, desert scene, computer monitors, heavy makeup, by Alphonse Mucha and Charlie Bowater, ((eyeshadow)), (coquettish), detailed, intricate +Negative prompt: ugly, fat, obese, chubby, (((deformed))), [blurry], bad anatomy, disfigured, poorly drawn face, mutation, mutated, (extra_limb), (ugly), (poorly drawn hands), messy drawing +Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 965400086, Size: 512x512, Model hash: 45dee52b +``` + + returns a dict with field values + """ + + res = {} + + prompt = "" + negative_prompt = "" + + done_with_prompt = False + + *lines, lastline = x.strip().split("\n") + if len(re_param.findall(lastline)) < 3: + lines.append(lastline) + lastline = '' + + for line in lines: + line = line.strip() + if line.startswith("Negative prompt:"): + done_with_prompt = True + line = line[16:].strip() + if done_with_prompt: + negative_prompt += ("" if negative_prompt == "" else "\n") + line + else: + prompt += ("" if prompt == "" else "\n") + line + + res["Prompt"] = prompt + res["Negative prompt"] = negative_prompt + + for k, v in re_param.findall(lastline): + if v[0] == '"' and v[-1] == '"': + v = unquote(v) + + m = re_imagesize.match(v) + if m is not None: + res[f"{k}-1"] = m.group(1) + res[f"{k}-2"] = m.group(2) + else: + res[k] = v + + # Missing CLIP skip means it was set to 1 (the default) + if "Clip skip" not in res: + res["Clip skip"] = "1" + + hypernet = res.get("Hypernet", None) + if hypernet is not None: + res["Prompt"] += f"""""" + + if "Hires resize-1" not in res: + res["Hires resize-1"] = 0 + res["Hires resize-2"] = 0 + + if "Hires sampler" not in res: + res["Hires sampler"] = "Use same sampler" + + if "Hires prompt" not in res: + res["Hires prompt"] = "" + + if "Hires negative prompt" not in res: + res["Hires negative prompt"] = "" + + restore_old_hires_fix_params(res) + + # Missing RNG means the default was set, which is GPU RNG + if "RNG" not in res: + res["RNG"] = "GPU" + + return res + + +settings_map = {} + + + +infotext_to_setting_name_mapping = [ + ('Clip skip', 'CLIP_stop_at_last_layers', ), + ('Conditional mask weight', 'inpainting_mask_weight'), + ('Model hash', 'sd_model_checkpoint'), + ('ENSD', 'eta_noise_seed_delta'), + ('Noise multiplier', 'initial_noise_multiplier'), + ('Eta', 'eta_ancestral'), + ('Eta DDIM', 'eta_ddim'), + ('Discard penultimate sigma', 'always_discard_next_to_last_sigma'), + ('UniPC variant', 'uni_pc_variant'), + ('UniPC skip type', 'uni_pc_skip_type'), + ('UniPC order', 'uni_pc_order'), + ('UniPC lower order final', 'uni_pc_lower_order_final'), + ('Token merging ratio', 'token_merging_ratio'), + ('Token merging ratio hr', 'token_merging_ratio_hr'), + ('RNG', 'randn_source'), + ('NGMS', 's_min_uncond'), +] + + +def create_override_settings_dict(text_pairs): + """creates processing's override_settings parameters from gradio's multiselect + + Example input: + ['Clip skip: 2', 'Model hash: e6e99610c4', 'ENSD: 31337'] + + Example output: + {'CLIP_stop_at_last_layers': 2, 'sd_model_checkpoint': 'e6e99610c4', 'eta_noise_seed_delta': 31337} + """ + + res = {} + + params = {} + for pair in text_pairs: + k, v = pair.split(":", maxsplit=1) + + params[k] = v.strip() + + for param_name, setting_name in infotext_to_setting_name_mapping: + value = params.get(param_name, None) + + if value is None: + continue + + res[setting_name] = shared.opts.cast_value(setting_name, value) + + return res + + +def connect_paste(button, paste_fields, input_comp, override_settings_component, tabname): + def paste_func(prompt): + if not prompt and not shared.cmd_opts.hide_ui_dir_config: + filename = os.path.join(data_path, "params.txt") + if os.path.exists(filename): + with open(filename, "r", encoding="utf8") as file: + prompt = file.read() + + params = parse_generation_parameters(prompt) + script_callbacks.infotext_pasted_callback(prompt, params) + res = [] + + for output, key in paste_fields: + if callable(key): + v = key(params) + else: + v = params.get(key, None) + + if v is None: + res.append(gr.update()) + elif isinstance(v, type_of_gr_update): + res.append(v) + else: + try: + valtype = type(output.value) + + if valtype == bool and v == "False": + val = False + else: + val = valtype(v) + + res.append(gr.update(value=val)) + except Exception: + res.append(gr.update()) + + return res + + if override_settings_component is not None: + def paste_settings(params): + vals = {} + + for param_name, setting_name in infotext_to_setting_name_mapping: + v = params.get(param_name, None) + if v is None: + continue + + if setting_name == "sd_model_checkpoint" and shared.opts.disable_weights_auto_swap: + continue + + v = shared.opts.cast_value(setting_name, v) + current_value = getattr(shared.opts, setting_name, None) + + if v == current_value: + continue + + vals[param_name] = v + + vals_pairs = [f"{k}: {v}" for k, v in vals.items()] + + return gr.Dropdown.update(value=vals_pairs, choices=vals_pairs, visible=len(vals_pairs) > 0) + + paste_fields = paste_fields + [(override_settings_component, paste_settings)] + + button.click( + fn=paste_func, + inputs=[input_comp], + outputs=[x[0] for x in paste_fields], + show_progress=False, + ) + button.click( + fn=None, + _js=f"recalculate_prompts_{tabname}", + inputs=[], + outputs=[], + show_progress=False, + ) + + diff --git a/modules/gfpgan_model.py b/modules/gfpgan_model.py new file mode 100644 index 0000000000000000000000000000000000000000..72ebe60d091fd088ca0a0ddb286ac1cb9e1f70e6 --- /dev/null +++ b/modules/gfpgan_model.py @@ -0,0 +1,116 @@ +import os +import sys +import traceback + +import facexlib +import gfpgan + +import modules.face_restoration +from modules import paths, shared, devices, modelloader + +model_dir = "GFPGAN" +user_path = None +model_path = os.path.join(paths.models_path, model_dir) +model_url = "https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.4.pth" +have_gfpgan = False +loaded_gfpgan_model = None + + +def gfpgann(): + global loaded_gfpgan_model + global model_path + if loaded_gfpgan_model is not None: + loaded_gfpgan_model.gfpgan.to(devices.device_gfpgan) + return loaded_gfpgan_model + + if gfpgan_constructor is None: + return None + + models = modelloader.load_models(model_path, model_url, user_path, ext_filter="GFPGAN") + if len(models) == 1 and "http" in models[0]: + model_file = models[0] + elif len(models) != 0: + latest_file = max(models, key=os.path.getctime) + model_file = latest_file + else: + print("Unable to load gfpgan model!") + return None + if hasattr(facexlib.detection.retinaface, 'device'): + facexlib.detection.retinaface.device = devices.device_gfpgan + model = gfpgan_constructor(model_path=model_file, upscale=1, arch='clean', channel_multiplier=2, bg_upsampler=None, device=devices.device_gfpgan) + loaded_gfpgan_model = model + + return model + + +def send_model_to(model, device): + model.gfpgan.to(device) + model.face_helper.face_det.to(device) + model.face_helper.face_parse.to(device) + + +def gfpgan_fix_faces(np_image): + model = gfpgann() + if model is None: + return np_image + + send_model_to(model, devices.device_gfpgan) + + np_image_bgr = np_image[:, :, ::-1] + cropped_faces, restored_faces, gfpgan_output_bgr = model.enhance(np_image_bgr, has_aligned=False, only_center_face=False, paste_back=True) + np_image = gfpgan_output_bgr[:, :, ::-1] + + model.face_helper.clean_all() + + if shared.opts.face_restoration_unload: + send_model_to(model, devices.cpu) + + return np_image + + +gfpgan_constructor = None + + +def setup_model(dirname): + global model_path + if not os.path.exists(model_path): + os.makedirs(model_path) + + try: + from gfpgan import GFPGANer + from facexlib import detection, parsing # noqa: F401 + global user_path + global have_gfpgan + global gfpgan_constructor + + load_file_from_url_orig = gfpgan.utils.load_file_from_url + facex_load_file_from_url_orig = facexlib.detection.load_file_from_url + facex_load_file_from_url_orig2 = facexlib.parsing.load_file_from_url + + def my_load_file_from_url(**kwargs): + return load_file_from_url_orig(**dict(kwargs, model_dir=model_path)) + + def facex_load_file_from_url(**kwargs): + return facex_load_file_from_url_orig(**dict(kwargs, save_dir=model_path, model_dir=None)) + + def facex_load_file_from_url2(**kwargs): + return facex_load_file_from_url_orig2(**dict(kwargs, save_dir=model_path, model_dir=None)) + + gfpgan.utils.load_file_from_url = my_load_file_from_url + facexlib.detection.load_file_from_url = facex_load_file_from_url + facexlib.parsing.load_file_from_url = facex_load_file_from_url2 + user_path = dirname + have_gfpgan = True + gfpgan_constructor = GFPGANer + + class FaceRestorerGFPGAN(modules.face_restoration.FaceRestoration): + def name(self): + return "GFPGAN" + + def restore(self, np_image): + return gfpgan_fix_faces(np_image) + + shared.face_restorers.append(FaceRestorerGFPGAN()) + except Exception: + print("Error setting up GFPGAN:", file=sys.stderr) + print(traceback.format_exc(), file=sys.stderr) diff --git a/modules/hashes.py b/modules/hashes.py new file mode 100644 index 0000000000000000000000000000000000000000..e17c730671f86d9607253cc704a06abba8a6d342 --- /dev/null +++ b/modules/hashes.py @@ -0,0 +1,108 @@ +import hashlib +import json +import os.path + +import filelock + +from modules import shared +from modules.paths import data_path + + +cache_filename = os.path.join(data_path, "cache.json") +cache_data = None + + +def dump_cache(): + with filelock.FileLock(f"{cache_filename}.lock"): + with open(cache_filename, "w", encoding="utf8") as file: + json.dump(cache_data, file, indent=4) + + +def cache(subsection): + global cache_data + + if cache_data is None: + with filelock.FileLock(f"{cache_filename}.lock"): + if not os.path.isfile(cache_filename): + cache_data = {} + else: + with open(cache_filename, "r", encoding="utf8") as file: + cache_data = json.load(file) + + s = cache_data.get(subsection, {}) + cache_data[subsection] = s + + return s + + +def calculate_sha256(filename): + hash_sha256 = hashlib.sha256() + blksize = 1024 * 1024 + + with open(filename, "rb") as f: + for chunk in iter(lambda: f.read(blksize), b""): + hash_sha256.update(chunk) + + return hash_sha256.hexdigest() + + +def sha256_from_cache(filename, title, use_addnet_hash=False): + hashes = cache("hashes-addnet") if use_addnet_hash else cache("hashes") + ondisk_mtime = os.path.getmtime(filename) + + if title not in hashes: + return None + + cached_sha256 = hashes[title].get("sha256", None) + cached_mtime = hashes[title].get("mtime", 0) + + if ondisk_mtime > cached_mtime or cached_sha256 is None: + return None + + return cached_sha256 + + +def sha256(filename, title, use_addnet_hash=False): + hashes = cache("hashes-addnet") if use_addnet_hash else cache("hashes") + + sha256_value = sha256_from_cache(filename, title, use_addnet_hash) + if sha256_value is not None: + return sha256_value + + if shared.cmd_opts.no_hashing: + return None + + print(f"Calculating sha256 for {filename}: ", end='') + if use_addnet_hash: + with open(filename, "rb") as file: + sha256_value = addnet_hash_safetensors(file) + else: + sha256_value = calculate_sha256(filename) + print(f"{sha256_value}") + + hashes[title] = { + "mtime": os.path.getmtime(filename), + "sha256": sha256_value, + } + + dump_cache() + + return sha256_value + + +def addnet_hash_safetensors(b): + """kohya-ss hash for safetensors from https://github.com/kohya-ss/sd-scripts/blob/main/library/train_util.py""" + hash_sha256 = hashlib.sha256() + blksize = 1024 * 1024 + + b.seek(0) + header = b.read(8) + n = int.from_bytes(header, "little") + + offset = n + 8 + b.seek(offset) + for chunk in iter(lambda: b.read(blksize), b""): + hash_sha256.update(chunk) + + return hash_sha256.hexdigest() + diff --git a/modules/hypernetworks/hypernetwork.py b/modules/hypernetworks/hypernetwork.py new file mode 100644 index 0000000000000000000000000000000000000000..6330f781e86e8a3004e03463e738caadde1b846e --- /dev/null +++ b/modules/hypernetworks/hypernetwork.py @@ -0,0 +1,810 @@ +import datetime +import glob +import html +import os +import sys +import traceback +import inspect + +import modules.textual_inversion.dataset +import torch +import tqdm +from einops import rearrange, repeat +from ldm.util import default +from modules import devices, processing, sd_models, shared, sd_samplers, hashes, sd_hijack_checkpoint +from modules.textual_inversion import textual_inversion, logging +from modules.textual_inversion.learn_schedule import LearnRateScheduler +from torch import einsum +from torch.nn.init import normal_, xavier_normal_, xavier_uniform_, kaiming_normal_, kaiming_uniform_, zeros_ + +from collections import deque +from statistics import stdev, mean + + +optimizer_dict = {optim_name : cls_obj for optim_name, cls_obj in inspect.getmembers(torch.optim, inspect.isclass) if optim_name != "Optimizer"} + +class HypernetworkModule(torch.nn.Module): + activation_dict = { + "linear": torch.nn.Identity, + "relu": torch.nn.ReLU, + "leakyrelu": torch.nn.LeakyReLU, + "elu": torch.nn.ELU, + "swish": torch.nn.Hardswish, + "tanh": torch.nn.Tanh, + "sigmoid": torch.nn.Sigmoid, + } + activation_dict.update({cls_name.lower(): cls_obj for cls_name, cls_obj in inspect.getmembers(torch.nn.modules.activation) if inspect.isclass(cls_obj) and cls_obj.__module__ == 'torch.nn.modules.activation'}) + + def __init__(self, dim, state_dict=None, layer_structure=None, activation_func=None, weight_init='Normal', + add_layer_norm=False, activate_output=False, dropout_structure=None): + super().__init__() + + self.multiplier = 1.0 + + assert layer_structure is not None, "layer_structure must not be None" + assert layer_structure[0] == 1, "Multiplier Sequence should start with size 1!" + assert layer_structure[-1] == 1, "Multiplier Sequence should end with size 1!" + + linears = [] + for i in range(len(layer_structure) - 1): + + # Add a fully-connected layer + linears.append(torch.nn.Linear(int(dim * layer_structure[i]), int(dim * layer_structure[i+1]))) + + # Add an activation func except last layer + if activation_func == "linear" or activation_func is None or (i >= len(layer_structure) - 2 and not activate_output): + pass + elif activation_func in self.activation_dict: + linears.append(self.activation_dict[activation_func]()) + else: + raise RuntimeError(f'hypernetwork uses an unsupported activation function: {activation_func}') + + # Add layer normalization + if add_layer_norm: + linears.append(torch.nn.LayerNorm(int(dim * layer_structure[i+1]))) + + # Everything should be now parsed into dropout structure, and applied here. + # Since we only have dropouts after layers, dropout structure should start with 0 and end with 0. + if dropout_structure is not None and dropout_structure[i+1] > 0: + assert 0 < dropout_structure[i+1] < 1, "Dropout probability should be 0 or float between 0 and 1!" + linears.append(torch.nn.Dropout(p=dropout_structure[i+1])) + # Code explanation : [1, 2, 1] -> dropout is missing when last_layer_dropout is false. [1, 2, 2, 1] -> [0, 0.3, 0, 0], when its True, [0, 0.3, 0.3, 0]. + + self.linear = torch.nn.Sequential(*linears) + + if state_dict is not None: + self.fix_old_state_dict(state_dict) + self.load_state_dict(state_dict) + else: + for layer in self.linear: + if type(layer) == torch.nn.Linear or type(layer) == torch.nn.LayerNorm: + w, b = layer.weight.data, layer.bias.data + if weight_init == "Normal" or type(layer) == torch.nn.LayerNorm: + normal_(w, mean=0.0, std=0.01) + normal_(b, mean=0.0, std=0) + elif weight_init == 'XavierUniform': + xavier_uniform_(w) + zeros_(b) + elif weight_init == 'XavierNormal': + xavier_normal_(w) + zeros_(b) + elif weight_init == 'KaimingUniform': + kaiming_uniform_(w, nonlinearity='leaky_relu' if 'leakyrelu' == activation_func else 'relu') + zeros_(b) + elif weight_init == 'KaimingNormal': + kaiming_normal_(w, nonlinearity='leaky_relu' if 'leakyrelu' == activation_func else 'relu') + zeros_(b) + else: + raise KeyError(f"Key {weight_init} is not defined as initialization!") + self.to(devices.device) + + def fix_old_state_dict(self, state_dict): + changes = { + 'linear1.bias': 'linear.0.bias', + 'linear1.weight': 'linear.0.weight', + 'linear2.bias': 'linear.1.bias', + 'linear2.weight': 'linear.1.weight', + } + + for fr, to in changes.items(): + x = state_dict.get(fr, None) + if x is None: + continue + + del state_dict[fr] + state_dict[to] = x + + def forward(self, x): + return x + self.linear(x) * (self.multiplier if not self.training else 1) + + def trainables(self): + layer_structure = [] + for layer in self.linear: + if type(layer) == torch.nn.Linear or type(layer) == torch.nn.LayerNorm: + layer_structure += [layer.weight, layer.bias] + return layer_structure + + +#param layer_structure : sequence used for length, use_dropout : controlling boolean, last_layer_dropout : for compatibility check. +def parse_dropout_structure(layer_structure, use_dropout, last_layer_dropout): + if layer_structure is None: + layer_structure = [1, 2, 1] + if not use_dropout: + return [0] * len(layer_structure) + dropout_values = [0] + dropout_values.extend([0.3] * (len(layer_structure) - 3)) + if last_layer_dropout: + dropout_values.append(0.3) + else: + dropout_values.append(0) + dropout_values.append(0) + return dropout_values + + +class Hypernetwork: + filename = None + name = None + + def __init__(self, name=None, enable_sizes=None, layer_structure=None, activation_func=None, weight_init=None, add_layer_norm=False, use_dropout=False, activate_output=False, **kwargs): + self.filename = None + self.name = name + self.layers = {} + self.step = 0 + self.sd_checkpoint = None + self.sd_checkpoint_name = None + self.layer_structure = layer_structure + self.activation_func = activation_func + self.weight_init = weight_init + self.add_layer_norm = add_layer_norm + self.use_dropout = use_dropout + self.activate_output = activate_output + self.last_layer_dropout = kwargs.get('last_layer_dropout', True) + self.dropout_structure = kwargs.get('dropout_structure', None) + if self.dropout_structure is None: + self.dropout_structure = parse_dropout_structure(self.layer_structure, self.use_dropout, self.last_layer_dropout) + self.optimizer_name = None + self.optimizer_state_dict = None + self.optional_info = None + + for size in enable_sizes or []: + self.layers[size] = ( + HypernetworkModule(size, None, self.layer_structure, self.activation_func, self.weight_init, + self.add_layer_norm, self.activate_output, dropout_structure=self.dropout_structure), + HypernetworkModule(size, None, self.layer_structure, self.activation_func, self.weight_init, + self.add_layer_norm, self.activate_output, dropout_structure=self.dropout_structure), + ) + self.eval() + + def weights(self): + res = [] + for layers in self.layers.values(): + for layer in layers: + res += layer.parameters() + return res + + def train(self, mode=True): + for layers in self.layers.values(): + for layer in layers: + layer.train(mode=mode) + for param in layer.parameters(): + param.requires_grad = mode + + def to(self, device): + for layers in self.layers.values(): + for layer in layers: + layer.to(device) + + return self + + def set_multiplier(self, multiplier): + for layers in self.layers.values(): + for layer in layers: + layer.multiplier = multiplier + + return self + + def eval(self): + for layers in self.layers.values(): + for layer in layers: + layer.eval() + for param in layer.parameters(): + param.requires_grad = False + + def save(self, filename): + state_dict = {} + optimizer_saved_dict = {} + + for k, v in self.layers.items(): + state_dict[k] = (v[0].state_dict(), v[1].state_dict()) + + state_dict['step'] = self.step + state_dict['name'] = self.name + state_dict['layer_structure'] = self.layer_structure + state_dict['activation_func'] = self.activation_func + state_dict['is_layer_norm'] = self.add_layer_norm + state_dict['weight_initialization'] = self.weight_init + state_dict['sd_checkpoint'] = self.sd_checkpoint + state_dict['sd_checkpoint_name'] = self.sd_checkpoint_name + state_dict['activate_output'] = self.activate_output + state_dict['use_dropout'] = self.use_dropout + state_dict['dropout_structure'] = self.dropout_structure + state_dict['last_layer_dropout'] = (self.dropout_structure[-2] != 0) if self.dropout_structure is not None else self.last_layer_dropout + state_dict['optional_info'] = self.optional_info if self.optional_info else None + + if self.optimizer_name is not None: + optimizer_saved_dict['optimizer_name'] = self.optimizer_name + + torch.save(state_dict, filename) + if shared.opts.save_optimizer_state and self.optimizer_state_dict: + optimizer_saved_dict['hash'] = self.shorthash() + optimizer_saved_dict['optimizer_state_dict'] = self.optimizer_state_dict + torch.save(optimizer_saved_dict, filename + '.optim') + + def load(self, filename): + self.filename = filename + if self.name is None: + self.name = os.path.splitext(os.path.basename(filename))[0] + + state_dict = torch.load(filename, map_location='cpu') + + self.layer_structure = state_dict.get('layer_structure', [1, 2, 1]) + self.optional_info = state_dict.get('optional_info', None) + self.activation_func = state_dict.get('activation_func', None) + self.weight_init = state_dict.get('weight_initialization', 'Normal') + self.add_layer_norm = state_dict.get('is_layer_norm', False) + self.dropout_structure = state_dict.get('dropout_structure', None) + self.use_dropout = True if self.dropout_structure is not None and any(self.dropout_structure) else state_dict.get('use_dropout', False) + self.activate_output = state_dict.get('activate_output', True) + self.last_layer_dropout = state_dict.get('last_layer_dropout', False) + # Dropout structure should have same length as layer structure, Every digits should be in [0,1), and last digit must be 0. + if self.dropout_structure is None: + self.dropout_structure = parse_dropout_structure(self.layer_structure, self.use_dropout, self.last_layer_dropout) + + if shared.opts.print_hypernet_extra: + if self.optional_info is not None: + print(f" INFO:\n {self.optional_info}\n") + + print(f" Layer structure: {self.layer_structure}") + print(f" Activation function: {self.activation_func}") + print(f" Weight initialization: {self.weight_init}") + print(f" Layer norm: {self.add_layer_norm}") + print(f" Dropout usage: {self.use_dropout}" ) + print(f" Activate last layer: {self.activate_output}") + print(f" Dropout structure: {self.dropout_structure}") + + optimizer_saved_dict = torch.load(self.filename + '.optim', map_location='cpu') if os.path.exists(self.filename + '.optim') else {} + + if self.shorthash() == optimizer_saved_dict.get('hash', None): + self.optimizer_state_dict = optimizer_saved_dict.get('optimizer_state_dict', None) + else: + self.optimizer_state_dict = None + if self.optimizer_state_dict: + self.optimizer_name = optimizer_saved_dict.get('optimizer_name', 'AdamW') + if shared.opts.print_hypernet_extra: + print("Loaded existing optimizer from checkpoint") + print(f"Optimizer name is {self.optimizer_name}") + else: + self.optimizer_name = "AdamW" + if shared.opts.print_hypernet_extra: + print("No saved optimizer exists in checkpoint") + + for size, sd in state_dict.items(): + if type(size) == int: + self.layers[size] = ( + HypernetworkModule(size, sd[0], self.layer_structure, self.activation_func, self.weight_init, + self.add_layer_norm, self.activate_output, self.dropout_structure), + HypernetworkModule(size, sd[1], self.layer_structure, self.activation_func, self.weight_init, + self.add_layer_norm, self.activate_output, self.dropout_structure), + ) + + self.name = state_dict.get('name', self.name) + self.step = state_dict.get('step', 0) + self.sd_checkpoint = state_dict.get('sd_checkpoint', None) + self.sd_checkpoint_name = state_dict.get('sd_checkpoint_name', None) + self.eval() + + def shorthash(self): + sha256 = hashes.sha256(self.filename, f'hypernet/{self.name}') + + return sha256[0:10] if sha256 else None + + +def list_hypernetworks(path): + res = {} + for filename in sorted(glob.iglob(os.path.join(path, '**/*.pt'), recursive=True), key=str.lower): + name = os.path.splitext(os.path.basename(filename))[0] + # Prevent a hypothetical "None.pt" from being listed. + if name != "None": + res[name] = filename + return res + + +def load_hypernetwork(name): + path = shared.hypernetworks.get(name, None) + + if path is None: + return None + + hypernetwork = Hypernetwork() + + try: + hypernetwork.load(path) + except Exception: + print(f"Error loading hypernetwork {path}", file=sys.stderr) + print(traceback.format_exc(), file=sys.stderr) + return None + + return hypernetwork + + +def load_hypernetworks(names, multipliers=None): + already_loaded = {} + + for hypernetwork in shared.loaded_hypernetworks: + if hypernetwork.name in names: + already_loaded[hypernetwork.name] = hypernetwork + + shared.loaded_hypernetworks.clear() + + for i, name in enumerate(names): + hypernetwork = already_loaded.get(name, None) + if hypernetwork is None: + hypernetwork = load_hypernetwork(name) + + if hypernetwork is None: + continue + + hypernetwork.set_multiplier(multipliers[i] if multipliers else 1.0) + shared.loaded_hypernetworks.append(hypernetwork) + + +def find_closest_hypernetwork_name(search: str): + if not search: + return None + search = search.lower() + applicable = [name for name in shared.hypernetworks if search in name.lower()] + if not applicable: + return None + applicable = sorted(applicable, key=lambda name: len(name)) + return applicable[0] + + +def apply_single_hypernetwork(hypernetwork, context_k, context_v, layer=None): + hypernetwork_layers = (hypernetwork.layers if hypernetwork is not None else {}).get(context_k.shape[2], None) + + if hypernetwork_layers is None: + return context_k, context_v + + if layer is not None: + layer.hyper_k = hypernetwork_layers[0] + layer.hyper_v = hypernetwork_layers[1] + + context_k = devices.cond_cast_unet(hypernetwork_layers[0](devices.cond_cast_float(context_k))) + context_v = devices.cond_cast_unet(hypernetwork_layers[1](devices.cond_cast_float(context_v))) + return context_k, context_v + + +def apply_hypernetworks(hypernetworks, context, layer=None): + context_k = context + context_v = context + for hypernetwork in hypernetworks: + context_k, context_v = apply_single_hypernetwork(hypernetwork, context_k, context_v, layer) + + return context_k, context_v + + +def attention_CrossAttention_forward(self, x, context=None, mask=None): + h = self.heads + + q = self.to_q(x) + context = default(context, x) + + context_k, context_v = apply_hypernetworks(shared.loaded_hypernetworks, context, self) + k = self.to_k(context_k) + v = self.to_v(context_v) + + q, k, v = (rearrange(t, 'b n (h d) -> (b h) n d', h=h) for t in (q, k, v)) + + sim = einsum('b i d, b j d -> b i j', q, k) * self.scale + + if mask is not None: + mask = rearrange(mask, 'b ... -> b (...)') + max_neg_value = -torch.finfo(sim.dtype).max + mask = repeat(mask, 'b j -> (b h) () j', h=h) + sim.masked_fill_(~mask, max_neg_value) + + # attention, what we cannot get enough of + attn = sim.softmax(dim=-1) + + out = einsum('b i j, b j d -> b i d', attn, v) + out = rearrange(out, '(b h) n d -> b n (h d)', h=h) + return self.to_out(out) + + +def stack_conds(conds): + if len(conds) == 1: + return torch.stack(conds) + + # same as in reconstruct_multicond_batch + token_count = max([x.shape[0] for x in conds]) + for i in range(len(conds)): + if conds[i].shape[0] != token_count: + last_vector = conds[i][-1:] + last_vector_repeated = last_vector.repeat([token_count - conds[i].shape[0], 1]) + conds[i] = torch.vstack([conds[i], last_vector_repeated]) + + return torch.stack(conds) + + +def statistics(data): + if len(data) < 2: + std = 0 + else: + std = stdev(data) + total_information = f"loss:{mean(data):.3f}" + u"\u00B1" + f"({std/ (len(data) ** 0.5):.3f})" + recent_data = data[-32:] + if len(recent_data) < 2: + std = 0 + else: + std = stdev(recent_data) + recent_information = f"recent 32 loss:{mean(recent_data):.3f}" + u"\u00B1" + f"({std / (len(recent_data) ** 0.5):.3f})" + return total_information, recent_information + + +def report_statistics(loss_info:dict): + keys = sorted(loss_info.keys(), key=lambda x: sum(loss_info[x]) / len(loss_info[x])) + for key in keys: + try: + print("Loss statistics for file " + key) + info, recent = statistics(list(loss_info[key])) + print(info) + print(recent) + except Exception as e: + print(e) + + +def create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure=None, activation_func=None, weight_init=None, add_layer_norm=False, use_dropout=False, dropout_structure=None): + # Remove illegal characters from name. + name = "".join( x for x in name if (x.isalnum() or x in "._- ")) + assert name, "Name cannot be empty!" + + fn = os.path.join(shared.cmd_opts.hypernetwork_dir, f"{name}.pt") + if not overwrite_old: + assert not os.path.exists(fn), f"file {fn} already exists" + + if type(layer_structure) == str: + layer_structure = [float(x.strip()) for x in layer_structure.split(",")] + + if use_dropout and dropout_structure and type(dropout_structure) == str: + dropout_structure = [float(x.strip()) for x in dropout_structure.split(",")] + else: + dropout_structure = [0] * len(layer_structure) + + hypernet = modules.hypernetworks.hypernetwork.Hypernetwork( + name=name, + enable_sizes=[int(x) for x in enable_sizes], + layer_structure=layer_structure, + activation_func=activation_func, + weight_init=weight_init, + add_layer_norm=add_layer_norm, + use_dropout=use_dropout, + dropout_structure=dropout_structure + ) + hypernet.save(fn) + + shared.reload_hypernetworks() + + +def train_hypernetwork(id_task, hypernetwork_name, learn_rate, batch_size, gradient_step, data_root, log_directory, training_width, training_height, varsize, steps, clip_grad_mode, clip_grad_value, shuffle_tags, tag_drop_out, latent_sampling_method, use_weight, create_image_every, save_hypernetwork_every, template_filename, preview_from_txt2img, preview_prompt, preview_negative_prompt, preview_steps, preview_sampler_index, preview_cfg_scale, preview_seed, preview_width, preview_height): + # images allows training previews to have infotext. Importing it at the top causes a circular import problem. + from modules import images + + save_hypernetwork_every = save_hypernetwork_every or 0 + create_image_every = create_image_every or 0 + template_file = textual_inversion.textual_inversion_templates.get(template_filename, None) + textual_inversion.validate_train_inputs(hypernetwork_name, learn_rate, batch_size, gradient_step, data_root, template_file, template_filename, steps, save_hypernetwork_every, create_image_every, log_directory, name="hypernetwork") + template_file = template_file.path + + path = shared.hypernetworks.get(hypernetwork_name, None) + hypernetwork = Hypernetwork() + hypernetwork.load(path) + shared.loaded_hypernetworks = [hypernetwork] + + shared.state.job = "train-hypernetwork" + shared.state.textinfo = "Initializing hypernetwork training..." + shared.state.job_count = steps + + hypernetwork_name = hypernetwork_name.rsplit('(', 1)[0] + filename = os.path.join(shared.cmd_opts.hypernetwork_dir, f'{hypernetwork_name}.pt') + + log_directory = os.path.join(log_directory, datetime.datetime.now().strftime("%Y-%m-%d"), hypernetwork_name) + unload = shared.opts.unload_models_when_training + + if save_hypernetwork_every > 0: + hypernetwork_dir = os.path.join(log_directory, "hypernetworks") + os.makedirs(hypernetwork_dir, exist_ok=True) + else: + hypernetwork_dir = None + + if create_image_every > 0: + images_dir = os.path.join(log_directory, "images") + os.makedirs(images_dir, exist_ok=True) + else: + images_dir = None + + checkpoint = sd_models.select_checkpoint() + + initial_step = hypernetwork.step or 0 + if initial_step >= steps: + shared.state.textinfo = "Model has already been trained beyond specified max steps" + return hypernetwork, filename + + scheduler = LearnRateScheduler(learn_rate, steps, initial_step) + + clip_grad = torch.nn.utils.clip_grad_value_ if clip_grad_mode == "value" else torch.nn.utils.clip_grad_norm_ if clip_grad_mode == "norm" else None + if clip_grad: + clip_grad_sched = LearnRateScheduler(clip_grad_value, steps, initial_step, verbose=False) + + if shared.opts.training_enable_tensorboard: + tensorboard_writer = textual_inversion.tensorboard_setup(log_directory) + + # dataset loading may take a while, so input validations and early returns should be done before this + shared.state.textinfo = f"Preparing dataset from {html.escape(data_root)}..." + + pin_memory = shared.opts.pin_memory + + ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=hypernetwork_name, model=shared.sd_model, cond_model=shared.sd_model.cond_stage_model, device=devices.device, template_file=template_file, include_cond=True, batch_size=batch_size, gradient_step=gradient_step, shuffle_tags=shuffle_tags, tag_drop_out=tag_drop_out, latent_sampling_method=latent_sampling_method, varsize=varsize, use_weight=use_weight) + + if shared.opts.save_training_settings_to_txt: + saved_params = dict( + model_name=checkpoint.model_name, model_hash=checkpoint.shorthash, num_of_dataset_images=len(ds), + **{field: getattr(hypernetwork, field) for field in ['layer_structure', 'activation_func', 'weight_init', 'add_layer_norm', 'use_dropout', ]} + ) + logging.save_settings_to_file(log_directory, {**saved_params, **locals()}) + + latent_sampling_method = ds.latent_sampling_method + + dl = modules.textual_inversion.dataset.PersonalizedDataLoader(ds, latent_sampling_method=latent_sampling_method, batch_size=ds.batch_size, pin_memory=pin_memory) + + old_parallel_processing_allowed = shared.parallel_processing_allowed + + if unload: + shared.parallel_processing_allowed = False + shared.sd_model.cond_stage_model.to(devices.cpu) + shared.sd_model.first_stage_model.to(devices.cpu) + + weights = hypernetwork.weights() + hypernetwork.train() + + # Here we use optimizer from saved HN, or we can specify as UI option. + if hypernetwork.optimizer_name in optimizer_dict: + optimizer = optimizer_dict[hypernetwork.optimizer_name](params=weights, lr=scheduler.learn_rate) + optimizer_name = hypernetwork.optimizer_name + else: + print(f"Optimizer type {hypernetwork.optimizer_name} is not defined!") + optimizer = torch.optim.AdamW(params=weights, lr=scheduler.learn_rate) + optimizer_name = 'AdamW' + + if hypernetwork.optimizer_state_dict: # This line must be changed if Optimizer type can be different from saved optimizer. + try: + optimizer.load_state_dict(hypernetwork.optimizer_state_dict) + except RuntimeError as e: + print("Cannot resume from saved optimizer!") + print(e) + + scaler = torch.cuda.amp.GradScaler() + + batch_size = ds.batch_size + gradient_step = ds.gradient_step + # n steps = batch_size * gradient_step * n image processed + steps_per_epoch = len(ds) // batch_size // gradient_step + max_steps_per_epoch = len(ds) // batch_size - (len(ds) // batch_size) % gradient_step + loss_step = 0 + _loss_step = 0 #internal + # size = len(ds.indexes) + # loss_dict = defaultdict(lambda : deque(maxlen = 1024)) + loss_logging = deque(maxlen=len(ds) * 3) # this should be configurable parameter, this is 3 * epoch(dataset size) + # losses = torch.zeros((size,)) + # previous_mean_losses = [0] + # previous_mean_loss = 0 + # print("Mean loss of {} elements".format(size)) + + steps_without_grad = 0 + + last_saved_file = "" + last_saved_image = "" + forced_filename = "" + + pbar = tqdm.tqdm(total=steps - initial_step) + try: + sd_hijack_checkpoint.add() + + for _ in range((steps-initial_step) * gradient_step): + if scheduler.finished: + break + if shared.state.interrupted: + break + for j, batch in enumerate(dl): + # works as a drop_last=True for gradient accumulation + if j == max_steps_per_epoch: + break + scheduler.apply(optimizer, hypernetwork.step) + if scheduler.finished: + break + if shared.state.interrupted: + break + + if clip_grad: + clip_grad_sched.step(hypernetwork.step) + + with devices.autocast(): + x = batch.latent_sample.to(devices.device, non_blocking=pin_memory) + if use_weight: + w = batch.weight.to(devices.device, non_blocking=pin_memory) + if tag_drop_out != 0 or shuffle_tags: + shared.sd_model.cond_stage_model.to(devices.device) + c = shared.sd_model.cond_stage_model(batch.cond_text).to(devices.device, non_blocking=pin_memory) + shared.sd_model.cond_stage_model.to(devices.cpu) + else: + c = stack_conds(batch.cond).to(devices.device, non_blocking=pin_memory) + if use_weight: + loss = shared.sd_model.weighted_forward(x, c, w)[0] / gradient_step + del w + else: + loss = shared.sd_model.forward(x, c)[0] / gradient_step + del x + del c + + _loss_step += loss.item() + scaler.scale(loss).backward() + + # go back until we reach gradient accumulation steps + if (j + 1) % gradient_step != 0: + continue + loss_logging.append(_loss_step) + if clip_grad: + clip_grad(weights, clip_grad_sched.learn_rate) + + scaler.step(optimizer) + scaler.update() + hypernetwork.step += 1 + pbar.update() + optimizer.zero_grad(set_to_none=True) + loss_step = _loss_step + _loss_step = 0 + + steps_done = hypernetwork.step + 1 + + epoch_num = hypernetwork.step // steps_per_epoch + epoch_step = hypernetwork.step % steps_per_epoch + + description = f"Training hypernetwork [Epoch {epoch_num}: {epoch_step+1}/{steps_per_epoch}]loss: {loss_step:.7f}" + pbar.set_description(description) + if hypernetwork_dir is not None and steps_done % save_hypernetwork_every == 0: + # Before saving, change name to match current checkpoint. + hypernetwork_name_every = f'{hypernetwork_name}-{steps_done}' + last_saved_file = os.path.join(hypernetwork_dir, f'{hypernetwork_name_every}.pt') + hypernetwork.optimizer_name = optimizer_name + if shared.opts.save_optimizer_state: + hypernetwork.optimizer_state_dict = optimizer.state_dict() + save_hypernetwork(hypernetwork, checkpoint, hypernetwork_name, last_saved_file) + hypernetwork.optimizer_state_dict = None # dereference it after saving, to save memory. + + + + if shared.opts.training_enable_tensorboard: + epoch_num = hypernetwork.step // len(ds) + epoch_step = hypernetwork.step - (epoch_num * len(ds)) + 1 + mean_loss = sum(loss_logging) / len(loss_logging) + textual_inversion.tensorboard_add(tensorboard_writer, loss=mean_loss, global_step=hypernetwork.step, step=epoch_step, learn_rate=scheduler.learn_rate, epoch_num=epoch_num) + + textual_inversion.write_loss(log_directory, "hypernetwork_loss.csv", hypernetwork.step, steps_per_epoch, { + "loss": f"{loss_step:.7f}", + "learn_rate": scheduler.learn_rate + }) + + if images_dir is not None and steps_done % create_image_every == 0: + forced_filename = f'{hypernetwork_name}-{steps_done}' + last_saved_image = os.path.join(images_dir, forced_filename) + hypernetwork.eval() + rng_state = torch.get_rng_state() + cuda_rng_state = None + if torch.cuda.is_available(): + cuda_rng_state = torch.cuda.get_rng_state_all() + shared.sd_model.cond_stage_model.to(devices.device) + shared.sd_model.first_stage_model.to(devices.device) + + p = processing.StableDiffusionProcessingTxt2Img( + sd_model=shared.sd_model, + do_not_save_grid=True, + do_not_save_samples=True, + ) + + p.disable_extra_networks = True + + if preview_from_txt2img: + p.prompt = preview_prompt + p.negative_prompt = preview_negative_prompt + p.steps = preview_steps + p.sampler_name = sd_samplers.samplers[preview_sampler_index].name + p.cfg_scale = preview_cfg_scale + p.seed = preview_seed + p.width = preview_width + p.height = preview_height + else: + p.prompt = batch.cond_text[0] + p.steps = 20 + p.width = training_width + p.height = training_height + + preview_text = p.prompt + + processed = processing.process_images(p) + image = processed.images[0] if len(processed.images) > 0 else None + + if unload: + shared.sd_model.cond_stage_model.to(devices.cpu) + shared.sd_model.first_stage_model.to(devices.cpu) + torch.set_rng_state(rng_state) + if torch.cuda.is_available(): + torch.cuda.set_rng_state_all(cuda_rng_state) + hypernetwork.train() + if image is not None: + shared.state.assign_current_image(image) + if shared.opts.training_enable_tensorboard and shared.opts.training_tensorboard_save_images: + textual_inversion.tensorboard_add_image(tensorboard_writer, + f"Validation at epoch {epoch_num}", image, + hypernetwork.step) + last_saved_image, last_text_info = images.save_image(image, images_dir, "", p.seed, p.prompt, shared.opts.samples_format, processed.infotexts[0], p=p, forced_filename=forced_filename, save_to_dirs=False) + last_saved_image += f", prompt: {preview_text}" + + shared.state.job_no = hypernetwork.step + + shared.state.textinfo = f""" +

+Loss: {loss_step:.7f}
+Step: {steps_done}
+Last prompt: {html.escape(batch.cond_text[0])}
+Last saved hypernetwork: {html.escape(last_saved_file)}
+Last saved image: {html.escape(last_saved_image)}
+

+""" + except Exception: + print(traceback.format_exc(), file=sys.stderr) + finally: + pbar.leave = False + pbar.close() + hypernetwork.eval() + #report_statistics(loss_dict) + sd_hijack_checkpoint.remove() + + + + filename = os.path.join(shared.cmd_opts.hypernetwork_dir, f'{hypernetwork_name}.pt') + hypernetwork.optimizer_name = optimizer_name + if shared.opts.save_optimizer_state: + hypernetwork.optimizer_state_dict = optimizer.state_dict() + save_hypernetwork(hypernetwork, checkpoint, hypernetwork_name, filename) + + del optimizer + hypernetwork.optimizer_state_dict = None # dereference it after saving, to save memory. + shared.sd_model.cond_stage_model.to(devices.device) + shared.sd_model.first_stage_model.to(devices.device) + shared.parallel_processing_allowed = old_parallel_processing_allowed + + return hypernetwork, filename + +def save_hypernetwork(hypernetwork, checkpoint, hypernetwork_name, filename): + old_hypernetwork_name = hypernetwork.name + old_sd_checkpoint = hypernetwork.sd_checkpoint if hasattr(hypernetwork, "sd_checkpoint") else None + old_sd_checkpoint_name = hypernetwork.sd_checkpoint_name if hasattr(hypernetwork, "sd_checkpoint_name") else None + try: + hypernetwork.sd_checkpoint = checkpoint.shorthash + hypernetwork.sd_checkpoint_name = checkpoint.model_name + hypernetwork.name = hypernetwork_name + hypernetwork.save(filename) + except: + hypernetwork.sd_checkpoint = old_sd_checkpoint + hypernetwork.sd_checkpoint_name = old_sd_checkpoint_name + hypernetwork.name = old_hypernetwork_name + raise diff --git a/modules/hypernetworks/ui.py b/modules/hypernetworks/ui.py new file mode 100644 index 0000000000000000000000000000000000000000..351910461dadbf3bfe027e542e0fddf896352d17 --- /dev/null +++ b/modules/hypernetworks/ui.py @@ -0,0 +1,38 @@ +import html + +import gradio as gr +import modules.hypernetworks.hypernetwork +from modules import devices, sd_hijack, shared + +not_available = ["hardswish", "multiheadattention"] +keys = [x for x in modules.hypernetworks.hypernetwork.HypernetworkModule.activation_dict if x not in not_available] + + +def create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure=None, activation_func=None, weight_init=None, add_layer_norm=False, use_dropout=False, dropout_structure=None): + filename = modules.hypernetworks.hypernetwork.create_hypernetwork(name, enable_sizes, overwrite_old, layer_structure, activation_func, weight_init, add_layer_norm, use_dropout, dropout_structure) + + return gr.Dropdown.update(choices=sorted(shared.hypernetworks)), f"Created: {filename}", "" + + +def train_hypernetwork(*args): + shared.loaded_hypernetworks = [] + + assert not shared.cmd_opts.lowvram, 'Training models with lowvram is not possible' + + try: + sd_hijack.undo_optimizations() + + hypernetwork, filename = modules.hypernetworks.hypernetwork.train_hypernetwork(*args) + + res = f""" +Training {'interrupted' if shared.state.interrupted else 'finished'} at {hypernetwork.step} steps. +Hypernetwork saved to {html.escape(filename)} +""" + return res, "" + except Exception: + raise + finally: + shared.sd_model.cond_stage_model.to(devices.device) + shared.sd_model.first_stage_model.to(devices.device) + sd_hijack.apply_optimizations() + diff --git a/modules/images.py b/modules/images.py new file mode 100644 index 0000000000000000000000000000000000000000..43886ebec11f3aefd97170e74bb6bab1662ab456 --- /dev/null +++ b/modules/images.py @@ -0,0 +1,715 @@ +import datetime +import sys +import traceback + +import pytz +import io +import math +import os +from collections import namedtuple +import re + +import numpy as np +import piexif +import piexif.helper +from PIL import Image, ImageFont, ImageDraw, PngImagePlugin +import string +import json +import hashlib + +from modules import sd_samplers, shared, script_callbacks, errors +from modules.paths_internal import roboto_ttf_file +from modules.shared import opts + +LANCZOS = (Image.Resampling.LANCZOS if hasattr(Image, 'Resampling') else Image.LANCZOS) + + +def get_font(fontsize: int): + try: + return ImageFont.truetype(opts.font or roboto_ttf_file, fontsize) + except Exception: + return ImageFont.truetype(roboto_ttf_file, fontsize) + + +def image_grid(imgs, batch_size=1, rows=None): + if rows is None: + if opts.n_rows > 0: + rows = opts.n_rows + elif opts.n_rows == 0: + rows = batch_size + elif opts.grid_prevent_empty_spots: + rows = math.floor(math.sqrt(len(imgs))) + while len(imgs) % rows != 0: + rows -= 1 + else: + rows = math.sqrt(len(imgs)) + rows = round(rows) + if rows > len(imgs): + rows = len(imgs) + + cols = math.ceil(len(imgs) / rows) + + params = script_callbacks.ImageGridLoopParams(imgs, cols, rows) + script_callbacks.image_grid_callback(params) + + w, h = imgs[0].size + grid = Image.new('RGB', size=(params.cols * w, params.rows * h), color='black') + + for i, img in enumerate(params.imgs): + grid.paste(img, box=(i % params.cols * w, i // params.cols * h)) + + return grid + + +Grid = namedtuple("Grid", ["tiles", "tile_w", "tile_h", "image_w", "image_h", "overlap"]) + + +def split_grid(image, tile_w=512, tile_h=512, overlap=64): + w = image.width + h = image.height + + non_overlap_width = tile_w - overlap + non_overlap_height = tile_h - overlap + + cols = math.ceil((w - overlap) / non_overlap_width) + rows = math.ceil((h - overlap) / non_overlap_height) + + dx = (w - tile_w) / (cols - 1) if cols > 1 else 0 + dy = (h - tile_h) / (rows - 1) if rows > 1 else 0 + + grid = Grid([], tile_w, tile_h, w, h, overlap) + for row in range(rows): + row_images = [] + + y = int(row * dy) + + if y + tile_h >= h: + y = h - tile_h + + for col in range(cols): + x = int(col * dx) + + if x + tile_w >= w: + x = w - tile_w + + tile = image.crop((x, y, x + tile_w, y + tile_h)) + + row_images.append([x, tile_w, tile]) + + grid.tiles.append([y, tile_h, row_images]) + + return grid + + +def combine_grid(grid): + def make_mask_image(r): + r = r * 255 / grid.overlap + r = r.astype(np.uint8) + return Image.fromarray(r, 'L') + + mask_w = make_mask_image(np.arange(grid.overlap, dtype=np.float32).reshape((1, grid.overlap)).repeat(grid.tile_h, axis=0)) + mask_h = make_mask_image(np.arange(grid.overlap, dtype=np.float32).reshape((grid.overlap, 1)).repeat(grid.image_w, axis=1)) + + combined_image = Image.new("RGB", (grid.image_w, grid.image_h)) + for y, h, row in grid.tiles: + combined_row = Image.new("RGB", (grid.image_w, h)) + for x, w, tile in row: + if x == 0: + combined_row.paste(tile, (0, 0)) + continue + + combined_row.paste(tile.crop((0, 0, grid.overlap, h)), (x, 0), mask=mask_w) + combined_row.paste(tile.crop((grid.overlap, 0, w, h)), (x + grid.overlap, 0)) + + if y == 0: + combined_image.paste(combined_row, (0, 0)) + continue + + combined_image.paste(combined_row.crop((0, 0, combined_row.width, grid.overlap)), (0, y), mask=mask_h) + combined_image.paste(combined_row.crop((0, grid.overlap, combined_row.width, h)), (0, y + grid.overlap)) + + return combined_image + + +class GridAnnotation: + def __init__(self, text='', is_active=True): + self.text = text + self.is_active = is_active + self.size = None + + +def draw_grid_annotations(im, width, height, hor_texts, ver_texts, margin=0): + def wrap(drawing, text, font, line_length): + lines = [''] + for word in text.split(): + line = f'{lines[-1]} {word}'.strip() + if drawing.textlength(line, font=font) <= line_length: + lines[-1] = line + else: + lines.append(word) + return lines + + def draw_texts(drawing, draw_x, draw_y, lines, initial_fnt, initial_fontsize): + for line in lines: + fnt = initial_fnt + fontsize = initial_fontsize + while drawing.multiline_textsize(line.text, font=fnt)[0] > line.allowed_width and fontsize > 0: + fontsize -= 1 + fnt = get_font(fontsize) + drawing.multiline_text((draw_x, draw_y + line.size[1] / 2), line.text, font=fnt, fill=color_active if line.is_active else color_inactive, anchor="mm", align="center") + + if not line.is_active: + drawing.line((draw_x - line.size[0] // 2, draw_y + line.size[1] // 2, draw_x + line.size[0] // 2, draw_y + line.size[1] // 2), fill=color_inactive, width=4) + + draw_y += line.size[1] + line_spacing + + fontsize = (width + height) // 25 + line_spacing = fontsize // 2 + + fnt = get_font(fontsize) + + color_active = (0, 0, 0) + color_inactive = (153, 153, 153) + + pad_left = 0 if sum([sum([len(line.text) for line in lines]) for lines in ver_texts]) == 0 else width * 3 // 4 + + cols = im.width // width + rows = im.height // height + + assert cols == len(hor_texts), f'bad number of horizontal texts: {len(hor_texts)}; must be {cols}' + assert rows == len(ver_texts), f'bad number of vertical texts: {len(ver_texts)}; must be {rows}' + + calc_img = Image.new("RGB", (1, 1), "white") + calc_d = ImageDraw.Draw(calc_img) + + for texts, allowed_width in zip(hor_texts + ver_texts, [width] * len(hor_texts) + [pad_left] * len(ver_texts)): + items = [] + texts + texts.clear() + + for line in items: + wrapped = wrap(calc_d, line.text, fnt, allowed_width) + texts += [GridAnnotation(x, line.is_active) for x in wrapped] + + for line in texts: + bbox = calc_d.multiline_textbbox((0, 0), line.text, font=fnt) + line.size = (bbox[2] - bbox[0], bbox[3] - bbox[1]) + line.allowed_width = allowed_width + + hor_text_heights = [sum([line.size[1] + line_spacing for line in lines]) - line_spacing for lines in hor_texts] + ver_text_heights = [sum([line.size[1] + line_spacing for line in lines]) - line_spacing * len(lines) for lines in ver_texts] + + pad_top = 0 if sum(hor_text_heights) == 0 else max(hor_text_heights) + line_spacing * 2 + + result = Image.new("RGB", (im.width + pad_left + margin * (cols-1), im.height + pad_top + margin * (rows-1)), "white") + + for row in range(rows): + for col in range(cols): + cell = im.crop((width * col, height * row, width * (col+1), height * (row+1))) + result.paste(cell, (pad_left + (width + margin) * col, pad_top + (height + margin) * row)) + + d = ImageDraw.Draw(result) + + for col in range(cols): + x = pad_left + (width + margin) * col + width / 2 + y = pad_top / 2 - hor_text_heights[col] / 2 + + draw_texts(d, x, y, hor_texts[col], fnt, fontsize) + + for row in range(rows): + x = pad_left / 2 + y = pad_top + (height + margin) * row + height / 2 - ver_text_heights[row] / 2 + + draw_texts(d, x, y, ver_texts[row], fnt, fontsize) + + return result + + +def draw_prompt_matrix(im, width, height, all_prompts, margin=0): + prompts = all_prompts[1:] + boundary = math.ceil(len(prompts) / 2) + + prompts_horiz = prompts[:boundary] + prompts_vert = prompts[boundary:] + + hor_texts = [[GridAnnotation(x, is_active=pos & (1 << i) != 0) for i, x in enumerate(prompts_horiz)] for pos in range(1 << len(prompts_horiz))] + ver_texts = [[GridAnnotation(x, is_active=pos & (1 << i) != 0) for i, x in enumerate(prompts_vert)] for pos in range(1 << len(prompts_vert))] + + return draw_grid_annotations(im, width, height, hor_texts, ver_texts, margin) + + +def resize_image(resize_mode, im, width, height, upscaler_name=None): + """ + Resizes an image with the specified resize_mode, width, and height. + + Args: + resize_mode: The mode to use when resizing the image. + 0: Resize the image to the specified width and height. + 1: Resize the image to fill the specified width and height, maintaining the aspect ratio, and then center the image within the dimensions, cropping the excess. + 2: Resize the image to fit within the specified width and height, maintaining the aspect ratio, and then center the image within the dimensions, filling empty with data from image. + im: The image to resize. + width: The width to resize the image to. + height: The height to resize the image to. + upscaler_name: The name of the upscaler to use. If not provided, defaults to opts.upscaler_for_img2img. + """ + + upscaler_name = upscaler_name or opts.upscaler_for_img2img + + def resize(im, w, h): + if upscaler_name is None or upscaler_name == "None" or im.mode == 'L': + return im.resize((w, h), resample=LANCZOS) + + scale = max(w / im.width, h / im.height) + + if scale > 1.0: + upscalers = [x for x in shared.sd_upscalers if x.name == upscaler_name] + if len(upscalers) == 0: + upscaler = shared.sd_upscalers[0] + print(f"could not find upscaler named {upscaler_name or ''}, using {upscaler.name} as a fallback") + else: + upscaler = upscalers[0] + + im = upscaler.scaler.upscale(im, scale, upscaler.data_path) + + if im.width != w or im.height != h: + im = im.resize((w, h), resample=LANCZOS) + + return im + + if resize_mode == 0: + res = resize(im, width, height) + + elif resize_mode == 1: + ratio = width / height + src_ratio = im.width / im.height + + src_w = width if ratio > src_ratio else im.width * height // im.height + src_h = height if ratio <= src_ratio else im.height * width // im.width + + resized = resize(im, src_w, src_h) + res = Image.new("RGB", (width, height)) + res.paste(resized, box=(width // 2 - src_w // 2, height // 2 - src_h // 2)) + + else: + ratio = width / height + src_ratio = im.width / im.height + + src_w = width if ratio < src_ratio else im.width * height // im.height + src_h = height if ratio >= src_ratio else im.height * width // im.width + + resized = resize(im, src_w, src_h) + res = Image.new("RGB", (width, height)) + res.paste(resized, box=(width // 2 - src_w // 2, height // 2 - src_h // 2)) + + if ratio < src_ratio: + fill_height = height // 2 - src_h // 2 + res.paste(resized.resize((width, fill_height), box=(0, 0, width, 0)), box=(0, 0)) + res.paste(resized.resize((width, fill_height), box=(0, resized.height, width, resized.height)), box=(0, fill_height + src_h)) + elif ratio > src_ratio: + fill_width = width // 2 - src_w // 2 + res.paste(resized.resize((fill_width, height), box=(0, 0, 0, height)), box=(0, 0)) + res.paste(resized.resize((fill_width, height), box=(resized.width, 0, resized.width, height)), box=(fill_width + src_w, 0)) + + return res + + +invalid_filename_chars = '<>:"/\\|?*\n' +invalid_filename_prefix = ' ' +invalid_filename_postfix = ' .' +re_nonletters = re.compile(r'[\s' + string.punctuation + ']+') +re_pattern = re.compile(r"(.*?)(?:\[([^\[\]]+)\]|$)") +re_pattern_arg = re.compile(r"(.*)<([^>]*)>$") +max_filename_part_length = 128 +NOTHING_AND_SKIP_PREVIOUS_TEXT = object() + + +def sanitize_filename_part(text, replace_spaces=True): + if text is None: + return None + + if replace_spaces: + text = text.replace(' ', '_') + + text = text.translate({ord(x): '_' for x in invalid_filename_chars}) + text = text.lstrip(invalid_filename_prefix)[:max_filename_part_length] + text = text.rstrip(invalid_filename_postfix) + return text + + +class FilenameGenerator: + replacements = { + 'seed': lambda self: self.seed if self.seed is not None else '', + 'steps': lambda self: self.p and self.p.steps, + 'cfg': lambda self: self.p and self.p.cfg_scale, + 'width': lambda self: self.image.width, + 'height': lambda self: self.image.height, + 'styles': lambda self: self.p and sanitize_filename_part(", ".join([style for style in self.p.styles if not style == "None"]) or "None", replace_spaces=False), + 'sampler': lambda self: self.p and sanitize_filename_part(self.p.sampler_name, replace_spaces=False), + 'model_hash': lambda self: getattr(self.p, "sd_model_hash", shared.sd_model.sd_model_hash), + 'model_name': lambda self: sanitize_filename_part(shared.sd_model.sd_checkpoint_info.model_name, replace_spaces=False), + 'date': lambda self: datetime.datetime.now().strftime('%Y-%m-%d'), + 'datetime': lambda self, *args: self.datetime(*args), # accepts formats: [datetime], [datetime], [datetime