Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+assets/13.gif filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,10 @@
+.DS_Store
+*pyc
+.vscode
+__pycache__
+*.egg-info
+
+checkpoints
+results
+backup
+LOG

LICENSE

@@ -0,0 +1,201 @@
+                                 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 Tencent
+
+   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.

README.md

@@ -1,12 +1,257 @@
----
-title: Tooncrafter
-emoji: 🏃
-colorFrom: pink
-colorTo: pink
-sdk: gradio
-sdk_version: 4.32.1
-app_file: app.py
-pinned: false
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+## ___***ToonCrafter: Generative Cartoon Interpolation***___
+<!-- ![](./assets/logo_long.png#gh-light-mode-only){: width="50%"} -->
+<!-- ![](./assets/logo_long_dark.png#gh-dark-mode-only=100x20) -->
+<div align="center">
+
+
+
+</div>
+ 
+## 🔆 Introduction
+
+⚠️ Please check our [disclaimer](#disc) first.
+
+🤗 ToonCrafter can interpolate two cartoon images by leveraging the pre-trained image-to-video diffusion priors. Please check our project page and paper for more information. <br>
+
+
+
+
+
+
+
+### 1.1 Showcases (512x320)
+<table class="center">
+    <tr style="font-weight: bolder;text-align:center;">
+        <td>Input starting frame</td>
+        <td>Input ending frame</td>
+        <td>Generated video</td>
+    </tr>
+  <tr>
+  <td>
+    <img src=assets/72109_125.mp4_00-00.png width="250">
+  </td>
+  <td>
+    <img src=assets/72109_125.mp4_00-01.png width="250">
+  </td>
+  <td>
+    <img src=assets/00.gif width="250">
+  </td>
+  </tr>
+
+
+   <tr>
+  <td>
+    <img src=assets/Japan_v2_2_062266_s2_frame1.png width="250">
+  </td>
+  <td>
+    <img src=assets/Japan_v2_2_062266_s2_frame3.png width="250">
+  </td>
+  <td>
+    <img src=assets/03.gif width="250">
+  </td>
+  </tr>
+  <tr>
+  <td>
+    <img src=assets/Japan_v2_1_070321_s3_frame1.png width="250">
+  </td>
+  <td>
+    <img src=assets/Japan_v2_1_070321_s3_frame3.png width="250">
+  </td>
+  <td>
+    <img src=assets/02.gif width="250">
+  </td>
+  </tr> 
+  <tr>
+  <td>
+    <img src=assets/74302_1349_frame1.png width="250">
+  </td>
+  <td>
+    <img src=assets/74302_1349_frame3.png width="250">
+  </td>
+  <td>
+    <img src=assets/01.gif width="250">
+  </td>
+  </tr>
+</table>
+
+### 1.2 Sparse sketch guidance
+<table class="center">
+    <tr style="font-weight: bolder;text-align:center;">
+        <td>Input starting frame</td>
+        <td>Input ending frame</td>
+        <td>Input sketch guidance</td>
+        <td>Generated video</td>
+    </tr>
+  <tr>
+  <td>
+    <img src=assets/72105_388.mp4_00-00.png width="200">
+  </td>
+  <td>
+    <img src=assets/72105_388.mp4_00-01.png width="200">
+  </td>
+  <td>
+    <img src=assets/06.gif width="200">
+  </td>
+   <td>
+    <img src=assets/07.gif width="200">
+  </td>
+  </tr>
+
+  <tr>
+  <td>
+    <img src=assets/72110_255.mp4_00-00.png width="200">
+  </td>
+  <td>
+    <img src=assets/72110_255.mp4_00-01.png width="200">
+  </td>
+  <td>
+    <img src=assets/12.gif width="200">
+  </td>
+   <td>
+    <img src=assets/13.gif width="200">
+  </td>
+  </tr>
+
+
+</table>
+
+
+### 2. Applications
+#### 2.1 Cartoon Sketch Interpolation (see project page for more details)
+<table class="center">
+    <tr style="font-weight: bolder;text-align:center;">
+        <td>Input starting frame</td>
+        <td>Input ending frame</td>
+        <td>Generated video</td>
+    </tr>
+
+  <tr>
+  <td>
+    <img src=assets/frame0001_10.png width="250">
+  </td>
+  <td>
+    <img src=assets/frame0016_10.png width="250">
+  </td>
+  <td>
+    <img src=assets/10.gif width="250">
+  </td>
+  </tr>
+
+
+   <tr>
+  <td>
+    <img src=assets/frame0001_11.png width="250">
+  </td>
+  <td>
+    <img src=assets/frame0016_11.png width="250">
+  </td>
+  <td>
+    <img src=assets/11.gif width="250">
+  </td>
+  </tr>
+
+</table>
+
+
+#### 2.2 Reference-based Sketch Colorization
+<table class="center">
+    <tr style="font-weight: bolder;text-align:center;">
+        <td>Input sketch</td>
+        <td>Input reference</td>
+        <td>Colorization results</td>
+    </tr>
+    
+  <tr>
+  <td>
+    <img src=assets/04.gif width="250">
+  </td>
+  <td>
+    <img src=assets/frame0001_05.png width="250">
+  </td>
+  <td>
+    <img src=assets/05.gif width="250">
+  </td>
+  </tr>
+
+
+   <tr>
+  <td>
+    <img src=assets/08.gif width="250">
+  </td>
+  <td>
+    <img src=assets/frame0001_09.png width="250">
+  </td>
+  <td>
+    <img src=assets/09.gif width="250">
+  </td>
+  </tr>
+
+</table>
+
+
+
+
+
+
+
+## 📝 Changelog
+- [ ] Add sketch control and colorization function.
+- __[2024.05.29]__: 🔥🔥 Release code and model weights.
+- __[2024.05.28]__: Launch the project page and update the arXiv preprint.
+<br>
+
+
+## 🧰 Models
+
+|Model|Resolution|GPU Mem. & Inference Time (A100, ddim 50steps)|Checkpoint|
+|:---------|:---------|:--------|:--------|
+|ToonCrafter_512|320x512| TBD (`perframe_ae=True`)|[Hugging Face](https://huggingface.co/Doubiiu/ToonCrafter/blob/main/model.ckpt)|
+
+
+Currently, our ToonCrafter can support generating videos of up to 16 frames with a resolution of 512x320. The inference time can be reduced by using fewer DDIM steps.
+
+
+
+## ⚙️ Setup
+
+### Install Environment via Anaconda (Recommended)
+```bash
+conda create -n tooncrafter python=3.8.5
+conda activate tooncrafter
+pip install -r requirements.txt
+```
+
+
+## 💫 Inference
+### 1. Command line
+
+Download pretrained ToonCrafter_512 and put the `model.ckpt` in `checkpoints/tooncrafter_512_interp_v1/model.ckpt`.
+```bash
+  sh scripts/run.sh
+```
+
+
+### 2. Local Gradio demo
+
+Download the pretrained model and put it in the corresponding directory according to the previous guidelines.
+```bash
+  python gradio_app.py 
+```
+
+
+
+
+
+
+<!-- ## 🤝 Community Support -->
+
+
+
+<a name="disc"></a>
+## 📢 Disclaimer
+Calm down. Our framework opens up the era of generative cartoon interpolation, but due to the variaity of generative video prior, the success rate is not guaranteed.
+
+⚠️This is an open-source research exploration, instead of commercial products. It can't meet all your expectations.
+
+This project strives to impact the domain of AI-driven video generation positively. Users are granted the freedom to create videos using this tool, but they are expected to comply with local laws and utilize it responsibly. The developers do not assume any responsibility for potential misuse by users.
+****

configs/inference_512_v1.0.yaml

@@ -0,0 +1,103 @@
+model:
+  target: lvdm.models.ddpm3d.LatentVisualDiffusion
+  params:
+    rescale_betas_zero_snr: True
+    parameterization: "v"
+    linear_start: 0.00085
+    linear_end: 0.012
+    num_timesteps_cond: 1
+    timesteps: 1000
+    first_stage_key: video
+    cond_stage_key: caption
+    cond_stage_trainable: False
+    conditioning_key: hybrid
+    image_size: [40, 64]
+    channels: 4
+    scale_by_std: False
+    scale_factor: 0.18215
+    use_ema: False
+    uncond_type: 'empty_seq'
+    use_dynamic_rescale: true
+    base_scale: 0.7
+    fps_condition_type: 'fps'
+    perframe_ae: True
+    loop_video: true
+    unet_config:
+      target: lvdm.modules.networks.openaimodel3d.UNetModel
+      params:
+        in_channels: 8
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions:
+        - 4
+        - 2
+        - 1
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 4
+        - 4
+        dropout: 0.1
+        num_head_channels: 64
+        transformer_depth: 1
+        context_dim: 1024
+        use_linear: true
+        use_checkpoint: True
+        temporal_conv: True
+        temporal_attention: True
+        temporal_selfatt_only: true
+        use_relative_position: false
+        use_causal_attention: False
+        temporal_length: 16
+        addition_attention: true
+        image_cross_attention: true
+        default_fs: 24
+        fs_condition: true
+
+    first_stage_config:
+      target: lvdm.models.autoencoder.AutoencoderKL_Dualref
+      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: lvdm.modules.encoders.condition.FrozenOpenCLIPEmbedder
+      params:
+        freeze: true
+        layer: "penultimate"
+
+    img_cond_stage_config:
+      target: lvdm.modules.encoders.condition.FrozenOpenCLIPImageEmbedderV2
+      params:
+        freeze: true
+    
+    image_proj_stage_config:
+      target: lvdm.modules.encoders.resampler.Resampler
+      params:
+        dim: 1024
+        depth: 4
+        dim_head: 64
+        heads: 12
+        num_queries: 16
+        embedding_dim: 1280
+        output_dim: 1024
+        ff_mult: 4
+        video_length: 16

configs/training_1024_v1.0/config.yaml

@@ -0,0 +1,166 @@
+model:
+  pretrained_checkpoint: checkpoints/dynamicrafter_1024_v1/model.ckpt
+  base_learning_rate: 1.0e-05
+  scale_lr: False
+  target: lvdm.models.ddpm3d.LatentVisualDiffusion
+  params:
+    rescale_betas_zero_snr: True
+    parameterization: "v"
+    linear_start: 0.00085
+    linear_end: 0.012
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: video
+    cond_stage_key: caption
+    cond_stage_trainable: False
+    image_proj_model_trainable: True
+    conditioning_key: hybrid
+    image_size: [72, 128]
+    channels: 4
+    scale_by_std: False
+    scale_factor: 0.18215
+    use_ema: False
+    uncond_prob: 0.05
+    uncond_type: 'empty_seq'
+    rand_cond_frame: true
+    use_dynamic_rescale: true
+    base_scale: 0.3
+    fps_condition_type: 'fps'
+    perframe_ae: True
+
+    unet_config:
+      target: lvdm.modules.networks.openaimodel3d.UNetModel
+      params:
+        in_channels: 8
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions:
+        - 4
+        - 2
+        - 1
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 4
+        - 4
+        dropout: 0.1
+        num_head_channels: 64
+        transformer_depth: 1
+        context_dim: 1024
+        use_linear: true
+        use_checkpoint: True
+        temporal_conv: True
+        temporal_attention: True
+        temporal_selfatt_only: true
+        use_relative_position: false
+        use_causal_attention: False
+        temporal_length: 16
+        addition_attention: true
+        image_cross_attention: true
+        default_fs: 10
+        fs_condition: true
+
+    first_stage_config:
+      target: lvdm.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: lvdm.modules.encoders.condition.FrozenOpenCLIPEmbedder
+      params:
+        freeze: true
+        layer: "penultimate"
+
+    img_cond_stage_config:
+      target: lvdm.modules.encoders.condition.FrozenOpenCLIPImageEmbedderV2
+      params:
+        freeze: true
+    
+    image_proj_stage_config:
+      target: lvdm.modules.encoders.resampler.Resampler
+      params:
+        dim: 1024
+        depth: 4
+        dim_head: 64
+        heads: 12
+        num_queries: 16
+        embedding_dim: 1280
+        output_dim: 1024
+        ff_mult: 4
+        video_length: 16
+
+data:
+  target: utils_data.DataModuleFromConfig
+  params:
+    batch_size: 1
+    num_workers: 12
+    wrap: false
+    train:
+      target: lvdm.data.webvid.WebVid
+      params:
+        data_dir: <WebVid10M DATA>
+        meta_path: <.csv FILE>
+        video_length: 16
+        frame_stride: 6
+        load_raw_resolution: true
+        resolution: [576, 1024]
+        spatial_transform: resize_center_crop
+        random_fs: true  ## if true, we uniformly sample fs with max_fs=frame_stride (above)
+
+lightning:
+  precision: 16
+  # strategy: deepspeed_stage_2
+  trainer:
+    benchmark: True
+    accumulate_grad_batches: 2
+    max_steps: 100000
+    # logger
+    log_every_n_steps: 50
+    # val
+    val_check_interval: 0.5
+    gradient_clip_algorithm: 'norm'
+    gradient_clip_val: 0.5
+  callbacks:
+    model_checkpoint:
+      target: pytorch_lightning.callbacks.ModelCheckpoint
+      params:
+        every_n_train_steps: 9000 #1000
+        filename: "{epoch}-{step}"
+        save_weights_only: True
+    metrics_over_trainsteps_checkpoint:
+      target: pytorch_lightning.callbacks.ModelCheckpoint
+      params:
+        filename: '{epoch}-{step}'
+        save_weights_only: True
+        every_n_train_steps: 10000 #20000 # 3s/step*2w=
+    batch_logger:
+      target: callbacks.ImageLogger
+      params:
+        batch_frequency: 500
+        to_local: False
+        max_images: 8
+        log_images_kwargs:
+          ddim_steps: 50
+          unconditional_guidance_scale: 7.5
+          timestep_spacing: uniform_trailing
+          guidance_rescale: 0.7

configs/training_1024_v1.0/run.sh

@@ -0,0 +1,37 @@
+# NCCL configuration
+# export NCCL_DEBUG=INFO
+# export NCCL_IB_DISABLE=0
+# export NCCL_IB_GID_INDEX=3
+# export NCCL_NET_GDR_LEVEL=3
+# export NCCL_TOPO_FILE=/tmp/topo.txt
+
+# args
+name="training_1024_v1.0"
+config_file=configs/${name}/config.yaml
+
+# save root dir for logs, checkpoints, tensorboard record, etc.
+save_root="<YOUR_SAVE_ROOT_DIR>"
+
+mkdir -p $save_root/$name
+
+## run
+CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python3 -m torch.distributed.launch \
+--nproc_per_node=$HOST_GPU_NUM --nnodes=1 --master_addr=127.0.0.1 --master_port=12352 --node_rank=0 \
+./main/trainer.py \
+--base $config_file \
+--train \
+--name $name \
+--logdir $save_root \
+--devices $HOST_GPU_NUM \
+lightning.trainer.num_nodes=1
+
+## debugging
+# CUDA_VISIBLE_DEVICES=0,1,2,3 python3 -m torch.distributed.launch \
+# --nproc_per_node=4 --nnodes=1 --master_addr=127.0.0.1 --master_port=12352 --node_rank=0 \
+# ./main/trainer.py \
+# --base $config_file \
+# --train \
+# --name $name \
+# --logdir $save_root \
+# --devices 4 \
+# lightning.trainer.num_nodes=1

configs/training_512_v1.0/config.yaml

@@ -0,0 +1,166 @@
+model:
+  pretrained_checkpoint: checkpoints/dynamicrafter_512_v1/model.ckpt
+  base_learning_rate: 1.0e-05
+  scale_lr: False
+  target: lvdm.models.ddpm3d.LatentVisualDiffusion
+  params:
+    rescale_betas_zero_snr: True
+    parameterization: "v"
+    linear_start: 0.00085
+    linear_end: 0.012
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: video
+    cond_stage_key: caption
+    cond_stage_trainable: False
+    image_proj_model_trainable: True
+    conditioning_key: hybrid
+    image_size: [40, 64]
+    channels: 4
+    scale_by_std: False
+    scale_factor: 0.18215
+    use_ema: False
+    uncond_prob: 0.05
+    uncond_type: 'empty_seq'
+    rand_cond_frame: true
+    use_dynamic_rescale: true
+    base_scale: 0.7
+    fps_condition_type: 'fps'
+    perframe_ae: True
+
+    unet_config:
+      target: lvdm.modules.networks.openaimodel3d.UNetModel
+      params:
+        in_channels: 8
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions:
+        - 4
+        - 2
+        - 1
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 4
+        - 4
+        dropout: 0.1
+        num_head_channels: 64
+        transformer_depth: 1
+        context_dim: 1024
+        use_linear: true
+        use_checkpoint: True
+        temporal_conv: True
+        temporal_attention: True
+        temporal_selfatt_only: true
+        use_relative_position: false
+        use_causal_attention: False
+        temporal_length: 16
+        addition_attention: true
+        image_cross_attention: true
+        default_fs: 10
+        fs_condition: true
+
+    first_stage_config:
+      target: lvdm.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: lvdm.modules.encoders.condition.FrozenOpenCLIPEmbedder
+      params:
+        freeze: true
+        layer: "penultimate"
+
+    img_cond_stage_config:
+      target: lvdm.modules.encoders.condition.FrozenOpenCLIPImageEmbedderV2
+      params:
+        freeze: true
+    
+    image_proj_stage_config:
+      target: lvdm.modules.encoders.resampler.Resampler
+      params:
+        dim: 1024
+        depth: 4
+        dim_head: 64
+        heads: 12
+        num_queries: 16
+        embedding_dim: 1280
+        output_dim: 1024
+        ff_mult: 4
+        video_length: 16
+
+data:
+  target: utils_data.DataModuleFromConfig
+  params:
+    batch_size: 2
+    num_workers: 12
+    wrap: false
+    train:
+      target: lvdm.data.webvid.WebVid
+      params:
+        data_dir: <WebVid10M DATA>
+        meta_path: <.csv FILE>
+        video_length: 16
+        frame_stride: 6
+        load_raw_resolution: true
+        resolution: [320, 512]
+        spatial_transform: resize_center_crop
+        random_fs: true  ## if true, we uniformly sample fs with max_fs=frame_stride (above)
+
+lightning:
+  precision: 16
+  # strategy: deepspeed_stage_2
+  trainer:
+    benchmark: True
+    accumulate_grad_batches: 2
+    max_steps: 100000
+    # logger
+    log_every_n_steps: 50
+    # val
+    val_check_interval: 0.5
+    gradient_clip_algorithm: 'norm'
+    gradient_clip_val: 0.5
+  callbacks:
+    model_checkpoint:
+      target: pytorch_lightning.callbacks.ModelCheckpoint
+      params:
+        every_n_train_steps: 9000 #1000
+        filename: "{epoch}-{step}"
+        save_weights_only: True
+    metrics_over_trainsteps_checkpoint:
+      target: pytorch_lightning.callbacks.ModelCheckpoint
+      params:
+        filename: '{epoch}-{step}'
+        save_weights_only: True
+        every_n_train_steps: 10000 #20000 # 3s/step*2w=
+    batch_logger:
+      target: callbacks.ImageLogger
+      params:
+        batch_frequency: 500
+        to_local: False
+        max_images: 8
+        log_images_kwargs:
+          ddim_steps: 50
+          unconditional_guidance_scale: 7.5
+          timestep_spacing: uniform_trailing
+          guidance_rescale: 0.7

configs/training_512_v1.0/run.sh

@@ -0,0 +1,37 @@
+# NCCL configuration
+# export NCCL_DEBUG=INFO
+# export NCCL_IB_DISABLE=0
+# export NCCL_IB_GID_INDEX=3
+# export NCCL_NET_GDR_LEVEL=3
+# export NCCL_TOPO_FILE=/tmp/topo.txt
+
+# args
+name="training_512_v1.0"
+config_file=configs/${name}/config.yaml
+
+# save root dir for logs, checkpoints, tensorboard record, etc.
+save_root="<YOUR_SAVE_ROOT_DIR>"
+
+mkdir -p $save_root/$name
+
+## run
+CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python3 -m torch.distributed.launch \
+--nproc_per_node=$HOST_GPU_NUM --nnodes=1 --master_addr=127.0.0.1 --master_port=12352 --node_rank=0 \
+./main/trainer.py \
+--base $config_file \
+--train \
+--name $name \
+--logdir $save_root \
+--devices $HOST_GPU_NUM \
+lightning.trainer.num_nodes=1
+
+## debugging
+# CUDA_VISIBLE_DEVICES=0,1,2,3 python3 -m torch.distributed.launch \
+# --nproc_per_node=4 --nnodes=1 --master_addr=127.0.0.1 --master_port=12352 --node_rank=0 \
+# ./main/trainer.py \
+# --base $config_file \
+# --train \
+# --name $name \
+# --logdir $save_root \
+# --devices 4 \
+# lightning.trainer.num_nodes=1

gradio_app.py

@@ -0,0 +1,82 @@
+import os, argparse
+import sys
+import gradio as gr
+from scripts.gradio.i2v_test_application import Image2Video
+sys.path.insert(1, os.path.join(sys.path[0], 'lvdm'))
+
+
+i2v_examples_interp_512 = [
+    ['prompts/512_interp/74906_1462_frame1.png', 'walking man', 50, 7.5, 1.0, 10, 123, 'prompts/512_interp/74906_1462_frame3.png'],
+    ['prompts/512_interp/Japan_v2_2_062266_s2_frame1.png', 'an anime scene', 50, 7.5, 1.0, 10, 789, 'prompts/512_interp/Japan_v2_2_062266_s2_frame3.png'],
+    ['prompts/512_interp/Japan_v2_3_119235_s2_frame1.png', 'an anime scene', 50, 7.5, 1.0, 10, 123, 'prompts/512_interp/Japan_v2_3_119235_s2_frame3.png'],
+]
+
+
+
+
+def dynamicrafter_demo(result_dir='./tmp/', res=512):
+    if res == 1024:
+        resolution = '576_1024'
+        css = """#input_img {max-width: 1024px !important} #output_vid {max-width: 1024px; max-height:576px}"""
+    elif res == 512:
+        resolution = '320_512'
+        css = """#input_img {max-width: 512px !important} #output_vid {max-width: 512px; max-height: 320px} #input_img2 {max-width: 512px !important} #output_vid {max-width: 512px; max-height: 320px}"""
+    elif res == 256:
+        resolution = '256_256'
+        css = """#input_img {max-width: 256px !important} #output_vid {max-width: 256px; max-height: 256px}"""
+    else:
+        raise NotImplementedError(f"Unsupported resolution: {res}")
+    image2video = Image2Video(result_dir, resolution=resolution)
+    with gr.Blocks(analytics_enabled=False, css=css) as dynamicrafter_iface:
+
+
+
+        with gr.Tab(label='ToonCrafter_320x512'):
+            with gr.Column():
+                with gr.Row():
+                    with gr.Column():
+                        with gr.Row():
+                            i2v_input_image = gr.Image(label="Input Image1",elem_id="input_img")
+                        with gr.Row():
+                            i2v_input_text = gr.Text(label='Prompts')
+                        with gr.Row():
+                            i2v_seed = gr.Slider(label='Random Seed', minimum=0, maximum=50000, step=1, value=123)
+                            i2v_eta = gr.Slider(minimum=0.0, maximum=1.0, step=0.1, label='ETA', value=1.0, elem_id="i2v_eta")
+                            i2v_cfg_scale = gr.Slider(minimum=1.0, maximum=15.0, step=0.5, label='CFG Scale', value=7.5, elem_id="i2v_cfg_scale")
+                        with gr.Row():
+                            i2v_steps = gr.Slider(minimum=1, maximum=60, step=1, elem_id="i2v_steps", label="Sampling steps", value=50)
+                            i2v_motion = gr.Slider(minimum=5, maximum=30, step=1, elem_id="i2v_motion", label="FPS", value=10)
+                        i2v_end_btn = gr.Button("Generate")
+                    with gr.Column():
+                        with gr.Row():
+                            i2v_input_image2 = gr.Image(label="Input Image2",elem_id="input_img2")
+                        with gr.Row():
+                            i2v_output_video = gr.Video(label="Generated Video",elem_id="output_vid",autoplay=True,show_share_button=True)
+
+                gr.Examples(examples=i2v_examples_interp_512,
+                            inputs=[i2v_input_image, i2v_input_text, i2v_steps, i2v_cfg_scale, i2v_eta, i2v_motion, i2v_seed, i2v_input_image2],
+                            outputs=[i2v_output_video],
+                            fn = image2video.get_image,
+                            cache_examples=False,
+                )
+            i2v_end_btn.click(inputs=[i2v_input_image, i2v_input_text, i2v_steps, i2v_cfg_scale, i2v_eta, i2v_motion, i2v_seed, i2v_input_image2],
+                            outputs=[i2v_output_video],
+                            fn = image2video.get_image
+            )
+
+
+    return dynamicrafter_iface
+
+def get_parser():
+    parser = argparse.ArgumentParser()
+    return parser
+
+if __name__ == "__main__":
+    parser = get_parser()
+    args = parser.parse_args()
+
+    result_dir = os.path.join('./', 'results')
+    dynamicrafter_iface = dynamicrafter_demo(result_dir)
+    dynamicrafter_iface.queue(max_size=12)
+    dynamicrafter_iface.launch(max_threads=1)
+    # dynamicrafter_iface.launch(server_name='0.0.0.0', server_port=80, max_threads=1)

lvdm/basics.py

@@ -0,0 +1,100 @@
+# adopted from
+# https://github.com/openai/improved-diffusion/blob/main/improved_diffusion/gaussian_diffusion.py
+# and
+# https://github.com/lucidrains/denoising-diffusion-pytorch/blob/7706bdfc6f527f58d33f84b7b522e61e6e3164b3/denoising_diffusion_pytorch/denoising_diffusion_pytorch.py
+# and
+# https://github.com/openai/guided-diffusion/blob/0ba878e517b276c45d1195eb29f6f5f72659a05b/guided_diffusion/nn.py
+#
+# thanks!
+
+import torch.nn as nn
+from utils.utils import instantiate_from_config
+
+
+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 zero_module(module):
+    """
+    Zero out the parameters of a module and return it.
+    """
+    for p in module.parameters():
+        p.detach().zero_()
+    return module
+
+def scale_module(module, scale):
+    """
+    Scale the parameters of a module and return it.
+    """
+    for p in module.parameters():
+        p.detach().mul_(scale)
+    return module
+
+
+def conv_nd(dims, *args, **kwargs):
+    """
+    Create a 1D, 2D, or 3D convolution module.
+    """
+    if dims == 1:
+        return nn.Conv1d(*args, **kwargs)
+    elif dims == 2:
+        return nn.Conv2d(*args, **kwargs)
+    elif dims == 3:
+        return nn.Conv3d(*args, **kwargs)
+    raise ValueError(f"unsupported dimensions: {dims}")
+
+
+def linear(*args, **kwargs):
+    """
+    Create a linear module.
+    """
+    return nn.Linear(*args, **kwargs)
+
+
+def avg_pool_nd(dims, *args, **kwargs):
+    """
+    Create a 1D, 2D, or 3D average pooling module.
+    """
+    if dims == 1:
+        return nn.AvgPool1d(*args, **kwargs)
+    elif dims == 2:
+        return nn.AvgPool2d(*args, **kwargs)
+    elif dims == 3:
+        return nn.AvgPool3d(*args, **kwargs)
+    raise ValueError(f"unsupported dimensions: {dims}")
+
+
+def nonlinearity(type='silu'):
+    if type == 'silu':
+        return nn.SiLU()
+    elif type == 'leaky_relu':
+        return nn.LeakyReLU()
+
+
+class GroupNormSpecific(nn.GroupNorm):
+    def forward(self, x):
+        return super().forward(x.float()).type(x.dtype)
+
+
+def normalization(channels, num_groups=32):
+    """
+    Make a standard normalization layer.
+    :param channels: number of input channels.
+    :return: an nn.Module for normalization.
+    """
+    return GroupNormSpecific(num_groups, channels)
+
+
+class HybridConditioner(nn.Module):
+
+    def __init__(self, c_concat_config, c_crossattn_config):
+        super().__init__()
+        self.concat_conditioner = instantiate_from_config(c_concat_config)
+        self.crossattn_conditioner = instantiate_from_config(c_crossattn_config)
+
+    def forward(self, c_concat, c_crossattn):
+        c_concat = self.concat_conditioner(c_concat)
+        c_crossattn = self.crossattn_conditioner(c_crossattn)
+        return {'c_concat': [c_concat], 'c_crossattn': [c_crossattn]}

lvdm/common.py

@@ -0,0 +1,94 @@
+import math
+from inspect import isfunction
+import torch
+from torch import nn
+import torch.distributed as dist
+
+
+def gather_data(data, return_np=True):
+    ''' gather data from multiple processes to one list '''
+    data_list = [torch.zeros_like(data) for _ in range(dist.get_world_size())]
+    dist.all_gather(data_list, data)  # gather not supported with NCCL
+    if return_np:
+        data_list = [data.cpu().numpy() for data in data_list]
+    return data_list
+
+def autocast(f):
+    def do_autocast(*args, **kwargs):
+        with torch.cuda.amp.autocast(enabled=True,
+                                     dtype=torch.get_autocast_gpu_dtype(),
+                                     cache_enabled=torch.is_autocast_cache_enabled()):
+            return f(*args, **kwargs)
+    return do_autocast
+
+
+def extract_into_tensor(a, t, x_shape):
+    b, *_ = t.shape
+    out = a.gather(-1, t)
+    return out.reshape(b, *((1,) * (len(x_shape) - 1)))
+
+
+def noise_like(shape, device, repeat=False):
+    repeat_noise = lambda: torch.randn((1, *shape[1:]), device=device).repeat(shape[0], *((1,) * (len(shape) - 1)))
+    noise = lambda: torch.randn(shape, device=device)
+    return repeat_noise() if repeat else noise()
+
+
+def default(val, d):
+    if exists(val):
+        return val
+    return d() if isfunction(d) else d
+
+def exists(val):
+    return val is not None
+
+def identity(*args, **kwargs):
+    return nn.Identity()
+
+def uniq(arr):
+    return{el: True for el in arr}.keys()
+
+def mean_flat(tensor):
+    """
+    Take the mean over all non-batch dimensions.
+    """
+    return tensor.mean(dim=list(range(1, len(tensor.shape))))
+
+def ismap(x):
+    if not isinstance(x, torch.Tensor):
+        return False
+    return (len(x.shape) == 4) and (x.shape[1] > 3)
+
+def isimage(x):
+    if not isinstance(x,torch.Tensor):
+        return False
+    return (len(x.shape) == 4) and (x.shape[1] == 3 or x.shape[1] == 1)
+
+def max_neg_value(t):
+    return -torch.finfo(t.dtype).max
+
+def shape_to_str(x):
+    shape_str = "x".join([str(x) for x in x.shape])
+    return shape_str
+
+def init_(tensor):
+    dim = tensor.shape[-1]
+    std = 1 / math.sqrt(dim)
+    tensor.uniform_(-std, std)
+    return tensor
+
+ckpt = torch.utils.checkpoint.checkpoint
+def checkpoint(func, inputs, params, flag):
+    """
+    Evaluate a function without caching intermediate activations, allowing for
+    reduced memory at the expense of extra compute in the backward pass.
+    :param func: the function to evaluate.
+    :param inputs: the argument sequence to pass to `func`.
+    :param params: a sequence of parameters `func` depends on but does not
+                   explicitly take as arguments.
+    :param flag: if False, disable gradient checkpointing.
+    """
+    if flag:
+        return ckpt(func, *inputs, use_reentrant=False)
+    else:
+        return func(*inputs)

lvdm/data/base.py

@@ -0,0 +1,23 @@
+from abc import abstractmethod
+from torch.utils.data import IterableDataset
+
+
+class Txt2ImgIterableBaseDataset(IterableDataset):
+    '''
+    Define an interface to make the IterableDatasets for text2img data chainable
+    '''
+    def __init__(self, num_records=0, valid_ids=None, size=256):
+        super().__init__()
+        self.num_records = num_records
+        self.valid_ids = valid_ids
+        self.sample_ids = valid_ids
+        self.size = size
+
+        print(f'{self.__class__.__name__} dataset contains {self.__len__()} examples.')
+
+    def __len__(self):
+        return self.num_records
+
+    @abstractmethod
+    def __iter__(self):
+        pass

lvdm/data/webvid.py

@@ -0,0 +1,202 @@
+import os
+import random
+from tqdm import tqdm
+import pandas as pd
+from decord import VideoReader, cpu
+
+import torch
+from torch.utils.data import Dataset
+from torch.utils.data import DataLoader
+from torchvision import transforms
+
+
+class WebVid(Dataset):
+    """
+    WebVid Dataset.
+    Assumes webvid data is structured as follows.
+    Webvid/
+        videos/
+            000001_000050/      ($page_dir)
+                1.mp4           (videoid.mp4)
+                ...
+                5000.mp4
+            ...
+    """
+    def __init__(self,
+                 meta_path,
+                 data_dir,
+                 subsample=None,
+                 video_length=16,
+                 resolution=[256, 512],
+                 frame_stride=1,
+                 frame_stride_min=1,
+                 spatial_transform=None,
+                 crop_resolution=None,
+                 fps_max=None,
+                 load_raw_resolution=False,
+                 fixed_fps=None,
+                 random_fs=False,
+                 ):
+        self.meta_path = meta_path
+        self.data_dir = data_dir
+        self.subsample = subsample
+        self.video_length = video_length
+        self.resolution = [resolution, resolution] if isinstance(resolution, int) else resolution
+        self.fps_max = fps_max
+        self.frame_stride = frame_stride
+        self.frame_stride_min = frame_stride_min
+        self.fixed_fps = fixed_fps
+        self.load_raw_resolution = load_raw_resolution
+        self.random_fs = random_fs
+        self._load_metadata()
+        if spatial_transform is not None:
+            if spatial_transform == "random_crop":
+                self.spatial_transform = transforms.RandomCrop(crop_resolution)
+            elif spatial_transform == "center_crop":
+                self.spatial_transform = transforms.Compose([
+                    transforms.CenterCrop(resolution),
+                    ])            
+            elif spatial_transform == "resize_center_crop":
+                # assert(self.resolution[0] == self.resolution[1])
+                self.spatial_transform = transforms.Compose([
+                    transforms.Resize(min(self.resolution)),
+                    transforms.CenterCrop(self.resolution),
+                    ])
+            elif spatial_transform == "resize":
+                self.spatial_transform = transforms.Resize(self.resolution)
+            else:
+                raise NotImplementedError
+        else:
+            self.spatial_transform = None
+                
+    def _load_metadata(self):
+        metadata = pd.read_csv(self.meta_path)
+        print(f'>>> {len(metadata)} data samples loaded.')
+        if self.subsample is not None:
+            metadata = metadata.sample(self.subsample, random_state=0)
+   
+        metadata['caption'] = metadata['name']
+        del metadata['name']
+        self.metadata = metadata
+        self.metadata.dropna(inplace=True)
+
+    def _get_video_path(self, sample):
+        rel_video_fp = os.path.join(sample['page_dir'], str(sample['videoid']) + '.mp4')
+        full_video_fp = os.path.join(self.data_dir, 'videos', rel_video_fp)
+        return full_video_fp
+    
+    def __getitem__(self, index):
+        if self.random_fs:
+            frame_stride = random.randint(self.frame_stride_min, self.frame_stride)
+        else:
+            frame_stride = self.frame_stride
+
+        ## get frames until success
+        while True:
+            index = index % len(self.metadata)
+            sample = self.metadata.iloc[index]
+            video_path = self._get_video_path(sample)
+            ## video_path should be in the format of "....../WebVid/videos/$page_dir/$videoid.mp4"
+            caption = sample['caption']
+
+            try:
+                if self.load_raw_resolution:
+                    video_reader = VideoReader(video_path, ctx=cpu(0))
+                else:
+                    video_reader = VideoReader(video_path, ctx=cpu(0), width=530, height=300)
+                if len(video_reader) < self.video_length:
+                    print(f"video length ({len(video_reader)}) is smaller than target length({self.video_length})")
+                    index += 1
+                    continue
+                else:
+                    pass
+            except:
+                index += 1
+                print(f"Load video failed! path = {video_path}")
+                continue
+            
+            fps_ori = video_reader.get_avg_fps()
+            if self.fixed_fps is not None:
+                frame_stride = int(frame_stride * (1.0 * fps_ori / self.fixed_fps))
+
+            ## to avoid extreme cases when fixed_fps is used
+            frame_stride = max(frame_stride, 1)
+            
+            ## get valid range (adapting case by case)
+            required_frame_num = frame_stride * (self.video_length-1) + 1
+            frame_num = len(video_reader)
+            if frame_num < required_frame_num:
+                ## drop extra samples if fixed fps is required
+                if self.fixed_fps is not None and frame_num < required_frame_num * 0.5:
+                    index += 1
+                    continue
+                else:
+                    frame_stride = frame_num // self.video_length
+                    required_frame_num = frame_stride * (self.video_length-1) + 1
+
+            ## select a random clip
+            random_range = frame_num - required_frame_num
+            start_idx = random.randint(0, random_range) if random_range > 0 else 0
+
+            ## calculate frame indices
+            frame_indices = [start_idx + frame_stride*i for i in range(self.video_length)]
+            try:
+                frames = video_reader.get_batch(frame_indices)
+                break
+            except:
+                print(f"Get frames failed! path = {video_path}; [max_ind vs frame_total:{max(frame_indices)} / {frame_num}]")
+                index += 1
+                continue
+        
+        ## process data
+        assert(frames.shape[0] == self.video_length),f'{len(frames)}, self.video_length={self.video_length}'
+        frames = torch.tensor(frames.asnumpy()).permute(3, 0, 1, 2).float() # [t,h,w,c] -> [c,t,h,w]
+        
+        if self.spatial_transform is not None:
+            frames = self.spatial_transform(frames)
+        
+        if self.resolution is not None:
+            assert (frames.shape[2], frames.shape[3]) == (self.resolution[0], self.resolution[1]), f'frames={frames.shape}, self.resolution={self.resolution}'
+        
+        ## turn frames tensors to [-1,1]
+        frames = (frames / 255 - 0.5) * 2
+        fps_clip = fps_ori // frame_stride
+        if self.fps_max is not None and fps_clip > self.fps_max:
+            fps_clip = self.fps_max
+
+        data = {'video': frames, 'caption': caption, 'path': video_path, 'fps': fps_clip, 'frame_stride': frame_stride}
+        return data
+    
+    def __len__(self):
+        return len(self.metadata)
+
+
+if __name__== "__main__":
+    meta_path = "" ## path to the meta file
+    data_dir = "" ## path to the data directory
+    save_dir = "" ## path to the save directory
+    dataset = WebVid(meta_path,
+                 data_dir,
+                 subsample=None,
+                 video_length=16,
+                 resolution=[256,448],
+                 frame_stride=4,
+                 spatial_transform="resize_center_crop",
+                 crop_resolution=None,
+                 fps_max=None,
+                 load_raw_resolution=True
+                 )
+    dataloader = DataLoader(dataset,
+                    batch_size=1,
+                    num_workers=0,
+                    shuffle=False)
+
+    
+    import sys
+    sys.path.insert(1, os.path.join(sys.path[0], '..', '..'))
+    from utils.save_video import tensor_to_mp4
+    for i, batch in tqdm(enumerate(dataloader), desc="Data Batch"):
+        video = batch['video']
+        name = batch['path'][0].split('videos/')[-1].replace('/','_')
+        tensor_to_mp4(video, save_dir+'/'+name, fps=8)
+

lvdm/distributions.py

@@ -0,0 +1,95 @@
+import torch
+import numpy as np
+
+
+class AbstractDistribution:
+    def sample(self):
+        raise NotImplementedError()
+
+    def mode(self):
+        raise NotImplementedError()
+
+
+class DiracDistribution(AbstractDistribution):
+    def __init__(self, value):
+        self.value = value
+
+    def sample(self):
+        return self.value
+
+    def mode(self):
+        return self.value
+
+
+class DiagonalGaussianDistribution(object):
+    def __init__(self, parameters, deterministic=False):
+        self.parameters = parameters
+        self.mean, self.logvar = torch.chunk(parameters, 2, dim=1)
+        self.logvar = torch.clamp(self.logvar, -30.0, 20.0)
+        self.deterministic = deterministic
+        self.std = torch.exp(0.5 * self.logvar)
+        self.var = torch.exp(self.logvar)
+        if self.deterministic:
+            self.var = self.std = torch.zeros_like(self.mean).to(device=self.parameters.device)
+
+    def sample(self, noise=None):
+        if noise is None:
+            noise = torch.randn(self.mean.shape)
+        
+        x = self.mean + self.std * noise.to(device=self.parameters.device)
+        return x
+
+    def kl(self, other=None):
+        if self.deterministic:
+            return torch.Tensor([0.])
+        else:
+            if other is None:
+                return 0.5 * torch.sum(torch.pow(self.mean, 2)
+                                       + self.var - 1.0 - self.logvar,
+                                       dim=[1, 2, 3])
+            else:
+                return 0.5 * torch.sum(
+                    torch.pow(self.mean - other.mean, 2) / other.var
+                    + self.var / other.var - 1.0 - self.logvar + other.logvar,
+                    dim=[1, 2, 3])
+
+    def nll(self, sample, dims=[1,2,3]):
+        if self.deterministic:
+            return torch.Tensor([0.])
+        logtwopi = np.log(2.0 * np.pi)
+        return 0.5 * torch.sum(
+            logtwopi + self.logvar + torch.pow(sample - self.mean, 2) / self.var,
+            dim=dims)
+
+    def mode(self):
+        return self.mean
+
+
+def normal_kl(mean1, logvar1, mean2, logvar2):
+    """
+    source: https://github.com/openai/guided-diffusion/blob/27c20a8fab9cb472df5d6bdd6c8d11c8f430b924/guided_diffusion/losses.py#L12
+    Compute the KL divergence between two gaussians.
+    Shapes are automatically broadcasted, so batches can be compared to
+    scalars, among other use cases.
+    """
+    tensor = None
+    for obj in (mean1, logvar1, mean2, logvar2):
+        if isinstance(obj, torch.Tensor):
+            tensor = obj
+            break
+    assert tensor is not None, "at least one argument must be a Tensor"
+
+    # Force variances to be Tensors. Broadcasting helps convert scalars to
+    # Tensors, but it does not work for torch.exp().
+    logvar1, logvar2 = [
+        x if isinstance(x, torch.Tensor) else torch.tensor(x).to(tensor)
+        for x in (logvar1, logvar2)
+    ]
+
+    return 0.5 * (
+        -1.0
+        + logvar2
+        - logvar1
+        + torch.exp(logvar1 - logvar2)
+        + ((mean1 - mean2) ** 2) * torch.exp(-logvar2)
+    )

lvdm/ema.py

@@ -0,0 +1,76 @@
+import torch
+from torch import nn
+
+
+class LitEma(nn.Module):
+    def __init__(self, model, decay=0.9999, use_num_upates=True):
+        super().__init__()
+        if decay < 0.0 or decay > 1.0:
+            raise ValueError('Decay must be between 0 and 1')
+
+        self.m_name2s_name = {}
+        self.register_buffer('decay', torch.tensor(decay, dtype=torch.float32))
+        self.register_buffer('num_updates', torch.tensor(0,dtype=torch.int) if use_num_upates
+                             else torch.tensor(-1,dtype=torch.int))
+
+        for name, p in model.named_parameters():
+            if p.requires_grad:
+                #remove as '.'-character is not allowed in buffers
+                s_name = name.replace('.','')
+                self.m_name2s_name.update({name:s_name})
+                self.register_buffer(s_name,p.clone().detach().data)
+
+        self.collected_params = []
+
+    def forward(self,model):
+        decay = self.decay
+
+        if self.num_updates >= 0:
+            self.num_updates += 1
+            decay = min(self.decay,(1 + self.num_updates) / (10 + self.num_updates))
+
+        one_minus_decay = 1.0 - decay
+
+        with torch.no_grad():
+            m_param = dict(model.named_parameters())
+            shadow_params = dict(self.named_buffers())
+
+            for key in m_param:
+                if m_param[key].requires_grad:
+                    sname = self.m_name2s_name[key]
+                    shadow_params[sname] = shadow_params[sname].type_as(m_param[key])
+                    shadow_params[sname].sub_(one_minus_decay * (shadow_params[sname] - m_param[key]))
+                else:
+                    assert not key in self.m_name2s_name
+
+    def copy_to(self, model):
+        m_param = dict(model.named_parameters())
+        shadow_params = dict(self.named_buffers())
+        for key in m_param:
+            if m_param[key].requires_grad:
+                m_param[key].data.copy_(shadow_params[self.m_name2s_name[key]].data)
+            else:
+                assert not key in self.m_name2s_name
+
+    def store(self, parameters):
+        """
+        Save the current parameters for restoring later.
+        Args:
+          parameters: Iterable of `torch.nn.Parameter`; the parameters to be
+            temporarily stored.
+        """
+        self.collected_params = [param.clone() for param in parameters]
+
+    def restore(self, parameters):
+        """
+        Restore the parameters stored with the `store` method.
+        Useful to validate the model with EMA parameters without affecting the
+        original optimization process. Store the parameters before the
+        `copy_to` method. After validation (or model saving), use this to
+        restore the former parameters.
+        Args:
+          parameters: Iterable of `torch.nn.Parameter`; the parameters to be
+            updated with the stored parameters.
+        """
+        for c_param, param in zip(self.collected_params, parameters):
+            param.data.copy_(c_param.data)

lvdm/models/autoencoder.py

@@ -0,0 +1,275 @@
+import os
+from contextlib import contextmanager
+import torch
+import numpy as np
+from einops import rearrange
+import torch.nn.functional as F
+import pytorch_lightning as pl
+from lvdm.modules.networks.ae_modules import Encoder, Decoder
+from lvdm.distributions import DiagonalGaussianDistribution
+from utils.utils import instantiate_from_config
+
+TIMESTEPS=16
+class AutoencoderKL(pl.LightningModule):
+    def __init__(self,
+                 ddconfig,
+                 lossconfig,
+                 embed_dim,
+                 ckpt_path=None,
+                 ignore_keys=[],
+                 image_key="image",
+                 colorize_nlabels=None,
+                 monitor=None,
+                 test=False,
+                 logdir=None,
+                 input_dim=4,
+                 test_args=None,
+                 additional_decode_keys=None,
+                 use_checkpoint=False,
+                 diff_boost_factor=3.0,
+                 ):
+        super().__init__()
+        self.image_key = image_key
+        self.encoder = Encoder(**ddconfig)
+        self.decoder = Decoder(**ddconfig)
+        self.loss = instantiate_from_config(lossconfig)
+        assert ddconfig["double_z"]
+        self.quant_conv = torch.nn.Conv2d(2*ddconfig["z_channels"], 2*embed_dim, 1)
+        self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)
+        self.embed_dim = embed_dim
+        self.input_dim = input_dim
+        self.test = test
+        self.test_args = test_args
+        self.logdir = logdir
+        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
+        if ckpt_path is not None:
+            self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys)
+        if self.test:
+            self.init_test()
+    
+    def init_test(self,):
+        self.test = True
+        save_dir = os.path.join(self.logdir, "test")
+        if 'ckpt' in self.test_args:
+            ckpt_name = os.path.basename(self.test_args.ckpt).split('.ckpt')[0] + f'_epoch{self._cur_epoch}'
+            self.root = os.path.join(save_dir, ckpt_name)
+        else:
+            self.root = save_dir
+        if 'test_subdir' in self.test_args:
+            self.root = os.path.join(save_dir, self.test_args.test_subdir)
+
+        self.root_zs = os.path.join(self.root, "zs")
+        self.root_dec = os.path.join(self.root, "reconstructions")
+        self.root_inputs = os.path.join(self.root, "inputs")
+        os.makedirs(self.root, exist_ok=True)
+
+        if self.test_args.save_z:
+            os.makedirs(self.root_zs, exist_ok=True)
+        if self.test_args.save_reconstruction:
+            os.makedirs(self.root_dec, exist_ok=True)
+        if self.test_args.save_input:
+            os.makedirs(self.root_inputs, exist_ok=True)
+        assert(self.test_args is not None)
+        self.test_maximum = getattr(self.test_args, 'test_maximum', None) 
+        self.count = 0
+        self.eval_metrics = {}
+        self.decodes = []
+        self.save_decode_samples = 2048
+
+    def init_from_ckpt(self, path, ignore_keys=list()):
+        sd = torch.load(path, map_location="cpu")
+        try:
+            self._cur_epoch = sd['epoch']
+            sd = sd["state_dict"]
+        except:
+            self._cur_epoch = 'null'
+        keys = list(sd.keys())
+        for k in keys:
+            for ik in ignore_keys:
+                if k.startswith(ik):
+                    print("Deleting key {} from state_dict.".format(k))
+                    del sd[k]
+        self.load_state_dict(sd, strict=False)
+        # self.load_state_dict(sd, strict=True)
+        print(f"Restored from {path}")
+
+    def encode(self, x, return_hidden_states=False, **kwargs):
+        if return_hidden_states:
+            h, hidden = self.encoder(x, return_hidden_states)
+            moments = self.quant_conv(h)
+            posterior = DiagonalGaussianDistribution(moments)
+            return posterior, hidden
+        else:
+            h = self.encoder(x)
+            moments = self.quant_conv(h)
+            posterior = DiagonalGaussianDistribution(moments)
+            return posterior
+
+    def decode(self, z, **kwargs):
+        if len(kwargs) == 0: ## use the original decoder in AutoencoderKL
+            z = self.post_quant_conv(z)
+        dec = self.decoder(z, **kwargs)  ##change for SVD decoder by adding **kwargs
+        return dec
+
+    def forward(self, input, sample_posterior=True, **additional_decode_kwargs):
+        input_tuple = (input, )
+        forward_temp = partial(self._forward, sample_posterior=sample_posterior, **additional_decode_kwargs)
+        return checkpoint(forward_temp, input_tuple, self.parameters(), self.use_checkpoint)
+        
+
+    def _forward(self, input, sample_posterior=True, **additional_decode_kwargs):
+        posterior = self.encode(input)
+        if sample_posterior:
+            z = posterior.sample()
+        else:
+            z = posterior.mode()
+        dec = self.decode(z, **additional_decode_kwargs)
+        ## print(input.shape, dec.shape) torch.Size([16, 3, 256, 256]) torch.Size([16, 3, 256, 256])
+        return dec, posterior
+
+    def get_input(self, batch, k):
+        x = batch[k]
+        if x.dim() == 5 and self.input_dim == 4:
+            b,c,t,h,w = x.shape
+            self.b = b
+            self.t = t 
+            x = rearrange(x, 'b c t h w -> (b t) c h w')
+
+        return x
+
+    def training_step(self, batch, batch_idx, optimizer_idx):
+        inputs = self.get_input(batch, self.image_key)
+        reconstructions, posterior = self(inputs)
+
+        if optimizer_idx == 0:
+            # train encoder+decoder+logvar
+            aeloss, log_dict_ae = self.loss(inputs, reconstructions, posterior, optimizer_idx, self.global_step,
+                                            last_layer=self.get_last_layer(), split="train")
+            self.log("aeloss", aeloss, prog_bar=True, logger=True, on_step=True, on_epoch=True)
+            self.log_dict(log_dict_ae, prog_bar=False, logger=True, on_step=True, on_epoch=False)
+            return aeloss
+
+        if optimizer_idx == 1:
+            # train the discriminator
+            discloss, log_dict_disc = self.loss(inputs, reconstructions, posterior, optimizer_idx, self.global_step,
+                                                last_layer=self.get_last_layer(), split="train")
+
+            self.log("discloss", discloss, prog_bar=True, logger=True, on_step=True, on_epoch=True)
+            self.log_dict(log_dict_disc, prog_bar=False, logger=True, on_step=True, on_epoch=False)
+            return discloss
+
+    def validation_step(self, batch, batch_idx):
+        inputs = self.get_input(batch, self.image_key)
+        reconstructions, posterior = self(inputs)
+        aeloss, log_dict_ae = self.loss(inputs, reconstructions, posterior, 0, self.global_step,
+                                        last_layer=self.get_last_layer(), split="val")
+
+        discloss, log_dict_disc = self.loss(inputs, reconstructions, posterior, 1, self.global_step,
+                                            last_layer=self.get_last_layer(), split="val")
+
+        self.log("val/rec_loss", log_dict_ae["val/rec_loss"])
+        self.log_dict(log_dict_ae)
+        self.log_dict(log_dict_disc)
+        return self.log_dict
+    
+    def configure_optimizers(self):
+        lr = self.learning_rate
+        opt_ae = torch.optim.Adam(list(self.encoder.parameters())+
+                                  list(self.decoder.parameters())+
+                                  list(self.quant_conv.parameters())+
+                                  list(self.post_quant_conv.parameters()),
+                                  lr=lr, betas=(0.5, 0.9))
+        opt_disc = torch.optim.Adam(self.loss.discriminator.parameters(),
+                                    lr=lr, betas=(0.5, 0.9))
+        return [opt_ae, opt_disc], []
+
+    def get_last_layer(self):
+        return self.decoder.conv_out.weight
+
+    @torch.no_grad()
+    def log_images(self, batch, only_inputs=False, **kwargs):
+        log = dict()
+        x = self.get_input(batch, self.image_key)
+        x = x.to(self.device)
+        if not only_inputs:
+            xrec, posterior = 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["samples"] = self.decode(torch.randn_like(posterior.sample()))
+            log["reconstructions"] = xrec
+        log["inputs"] = x
+        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 IdentityFirstStage(torch.nn.Module):
+    def __init__(self, *args, vq_interface=False, **kwargs):
+        self.vq_interface = vq_interface  # TODO: Should be true by default but check to not break older stuff
+        super().__init__()
+
+    def encode(self, x, *args, **kwargs):
+        return x
+
+    def decode(self, x, *args, **kwargs):
+        return x
+
+    def quantize(self, x, *args, **kwargs):
+        if self.vq_interface:
+            return x, None, [None, None, None]
+        return x
+
+    def forward(self, x, *args, **kwargs):
+        return x
+
+from lvdm.models.autoencoder_dualref import VideoDecoder
+class AutoencoderKL_Dualref(AutoencoderKL):
+    def __init__(self,
+                 ddconfig,
+                 lossconfig,
+                 embed_dim,
+                 ckpt_path=None,
+                 ignore_keys=[],
+                 image_key="image",
+                 colorize_nlabels=None,
+                 monitor=None,
+                 test=False,
+                 logdir=None,
+                 input_dim=4,
+                 test_args=None,
+                 additional_decode_keys=None,
+                 use_checkpoint=False,
+                 diff_boost_factor=3.0,
+                 ):
+        super().__init__(ddconfig, lossconfig, embed_dim, ckpt_path, ignore_keys, image_key, colorize_nlabels, monitor, test, logdir, input_dim, test_args, additional_decode_keys, use_checkpoint, diff_boost_factor)
+        self.decoder = VideoDecoder(**ddconfig)
+
+    def _forward(self, input, sample_posterior=True, **additional_decode_kwargs):
+        posterior, hidden_states = self.encode(input, return_hidden_states=True)
+
+        hidden_states_first_last = []
+        ### use only the first and last hidden states
+        for hid in hidden_states:
+            hid = rearrange(hid, '(b t) c h w -> b c t h w', t=TIMESTEPS)
+            hid_new = torch.cat([hid[:, :, 0:1], hid[:, :, -1:]], dim=2)
+            hidden_states_first_last.append(hid_new)
+
+        if sample_posterior:
+            z = posterior.sample()
+        else:
+            z = posterior.mode()
+        dec = self.decode(z, ref_context=hidden_states_first_last, **additional_decode_kwargs)
+        ## print(input.shape, dec.shape) torch.Size([16, 3, 256, 256]) torch.Size([16, 3, 256, 256])
+        return dec, posterior

lvdm/models/autoencoder_dualref.py

@@ -0,0 +1,1177 @@
+#### https://github.com/Stability-AI/generative-models
+from einops import rearrange, repeat
+import logging
+from typing import Any, Callable, Optional, Iterable, Union
+
+import numpy as np
+import torch
+import torch.nn as nn
+from packaging import version
+logpy = logging.getLogger(__name__)
+
+try:
+    import xformers
+    import xformers.ops
+
+    XFORMERS_IS_AVAILABLE = True
+except:
+    XFORMERS_IS_AVAILABLE = False
+    logpy.warning("no module 'xformers'. Processing without...")
+
+from lvdm.modules.attention_svd import LinearAttention, MemoryEfficientCrossAttention
+
+
+def nonlinearity(x):
+    # swish
+    return x * torch.sigmoid(x)
+
+
+def Normalize(in_channels, num_groups=32):
+    return torch.nn.GroupNorm(
+        num_groups=num_groups, num_channels=in_channels, eps=1e-6, affine=True
+    )
+
+
+class ResnetBlock(nn.Module):
+    def __init__(
+        self,
+        *,
+        in_channels,
+        out_channels=None,
+        conv_shortcut=False,
+        dropout,
+        temb_channels=512,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        out_channels = in_channels if out_channels is None else out_channels
+        self.out_channels = out_channels
+        self.use_conv_shortcut = conv_shortcut
+
+        self.norm1 = Normalize(in_channels)
+        self.conv1 = torch.nn.Conv2d(
+            in_channels, out_channels, kernel_size=3, stride=1, padding=1
+        )
+        if temb_channels > 0:
+            self.temb_proj = torch.nn.Linear(temb_channels, out_channels)
+        self.norm2 = Normalize(out_channels)
+        self.dropout = torch.nn.Dropout(dropout)
+        self.conv2 = torch.nn.Conv2d(
+            out_channels, out_channels, kernel_size=3, stride=1, padding=1
+        )
+        if self.in_channels != self.out_channels:
+            if self.use_conv_shortcut:
+                self.conv_shortcut = torch.nn.Conv2d(
+                    in_channels, out_channels, kernel_size=3, stride=1, padding=1
+                )
+            else:
+                self.nin_shortcut = torch.nn.Conv2d(
+                    in_channels, out_channels, kernel_size=1, stride=1, padding=0
+                )
+
+    def forward(self, x, temb):
+        h = x
+        h = self.norm1(h)
+        h = nonlinearity(h)
+        h = self.conv1(h)
+
+        if temb is not None:
+            h = h + self.temb_proj(nonlinearity(temb))[:, :, None, None]
+
+        h = self.norm2(h)
+        h = nonlinearity(h)
+        h = self.dropout(h)
+        h = self.conv2(h)
+
+        if self.in_channels != self.out_channels:
+            if self.use_conv_shortcut:
+                x = self.conv_shortcut(x)
+            else:
+                x = self.nin_shortcut(x)
+
+        return x + h
+
+
+class LinAttnBlock(LinearAttention):
+    """to match AttnBlock usage"""
+
+    def __init__(self, in_channels):
+        super().__init__(dim=in_channels, heads=1, dim_head=in_channels)
+
+
+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 attention(self, h_: torch.Tensor) -> torch.Tensor:
+        h_ = self.norm(h_)
+        q = self.q(h_)
+        k = self.k(h_)
+        v = self.v(h_)
+
+        b, c, h, w = q.shape
+        q, k, v = map(
+            lambda x: rearrange(x, "b c h w -> b 1 (h w) c").contiguous(), (q, k, v)
+        )
+        h_ = torch.nn.functional.scaled_dot_product_attention(
+            q, k, v
+        )  # scale is dim ** -0.5 per default
+        # compute attention
+
+        return rearrange(h_, "b 1 (h w) c -> b c h w", h=h, w=w, c=c, b=b)
+
+    def forward(self, x, **kwargs):
+        h_ = x
+        h_ = self.attention(h_)
+        h_ = self.proj_out(h_)
+        return x + h_
+
+
+class MemoryEfficientAttnBlock(nn.Module):
+    """
+    Uses xformers efficient implementation,
+    see https://github.com/MatthieuTPHR/diffusers/blob/d80b531ff8060ec1ea982b65a1b8df70f73aa67c/src/diffusers/models/attention.py#L223
+    Note: this is a single-head self-attention operation
+    """
+
+    #
+    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
+        )
+        self.attention_op: Optional[Any] = None
+
+    def attention(self, h_: torch.Tensor) -> torch.Tensor:
+        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, k, v = map(lambda x: rearrange(x, "b c h w -> b (h w) c"), (q, k, v))
+
+        q, k, v = map(
+            lambda t: t.unsqueeze(3)
+            .reshape(B, t.shape[1], 1, C)
+            .permute(0, 2, 1, 3)
+            .reshape(B * 1, t.shape[1], C)
+            .contiguous(),
+            (q, k, v),
+        )
+        out = xformers.ops.memory_efficient_attention(
+            q, k, v, attn_bias=None, op=self.attention_op
+        )
+
+        out = (
+            out.unsqueeze(0)
+            .reshape(B, 1, out.shape[1], C)
+            .permute(0, 2, 1, 3)
+            .reshape(B, out.shape[1], C)
+        )
+        return rearrange(out, "b (h w) c -> b c h w", b=B, h=H, w=W, c=C)
+
+    def forward(self, x, **kwargs):
+        h_ = x
+        h_ = self.attention(h_)
+        h_ = self.proj_out(h_)
+        return x + h_
+
+
+class MemoryEfficientCrossAttentionWrapper(MemoryEfficientCrossAttention):
+    def forward(self, x, context=None, mask=None, **unused_kwargs):
+        b, c, h, w = x.shape
+        x = rearrange(x, "b c h w -> b (h w) c")
+        out = super().forward(x, context=context, mask=mask)
+        out = rearrange(out, "b (h w) c -> b c h w", h=h, w=w, c=c)
+        return x + out
+
+
+def make_attn(in_channels, attn_type="vanilla", attn_kwargs=None):
+    assert attn_type in [
+        "vanilla",
+        "vanilla-xformers",
+        "memory-efficient-cross-attn",
+        "linear",
+        "none",
+        "memory-efficient-cross-attn-fusion",
+    ], f"attn_type {attn_type} unknown"
+    if (
+        version.parse(torch.__version__) < version.parse("2.0.0")
+        and attn_type != "none"
+    ):
+        assert XFORMERS_IS_AVAILABLE, (
+            f"We do not support vanilla attention in {torch.__version__} anymore, "
+            f"as it is too expensive. Please install xformers via e.g. 'pip install xformers==0.0.16'"
+        )
+        # attn_type = "vanilla-xformers"
+    logpy.info(f"making attention of type '{attn_type}' with {in_channels} in_channels")
+    if attn_type == "vanilla":
+        assert attn_kwargs is None
+        return AttnBlock(in_channels)
+    elif attn_type == "vanilla-xformers":
+        logpy.info(
+            f"building MemoryEfficientAttnBlock with {in_channels} in_channels..."
+        )
+        return MemoryEfficientAttnBlock(in_channels)
+    elif attn_type == "memory-efficient-cross-attn":
+        attn_kwargs["query_dim"] = in_channels
+        return MemoryEfficientCrossAttentionWrapper(**attn_kwargs)
+    elif attn_type == "memory-efficient-cross-attn-fusion":
+        attn_kwargs["query_dim"] = in_channels
+        return MemoryEfficientCrossAttentionWrapperFusion(**attn_kwargs)
+    elif attn_type == "none":
+        return nn.Identity(in_channels)
+    else:
+        return LinAttnBlock(in_channels)
+
+class MemoryEfficientCrossAttentionWrapperFusion(MemoryEfficientCrossAttention):
+    # print('x.shape: ',x.shape, 'context.shape: ',context.shape) ##torch.Size([8, 128, 256, 256]) torch.Size([1, 128, 2, 256, 256])
+    def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0, **kwargs):
+        super().__init__(query_dim, context_dim, heads, dim_head, dropout, **kwargs)
+        self.norm = Normalize(query_dim)
+        nn.init.zeros_(self.to_out[0].weight)
+        nn.init.zeros_(self.to_out[0].bias)
+
+    def forward(self, x, context=None, mask=None):
+        if self.training:
+            return checkpoint(self._forward, x, context, mask, use_reentrant=False)
+        else:
+            return self._forward(x, context, mask)
+
+    def _forward(
+        self,
+        x,
+        context=None,
+        mask=None,
+    ):
+        bt, c, h, w = x.shape
+        h_ = self.norm(x)
+        h_ = rearrange(h_, "b c h w -> b (h w) c")
+        q = self.to_q(h_)
+
+
+        b, c, l, h, w = context.shape
+        context = rearrange(context, "b c l h w -> (b l) (h w) c")
+        k = self.to_k(context)
+        v = self.to_v(context)
+        k = rearrange(k, "(b l) d c -> b l d c", l=l)
+        k = torch.cat([k[:, [0] * (bt//b)], k[:, [1]*(bt//b)]], dim=2)
+        k = rearrange(k, "b l d c -> (b l) d c")
+
+        v = rearrange(v, "(b l) d c -> b l d c", l=l)
+        v = torch.cat([v[:, [0] * (bt//b)], v[:, [1]*(bt//b)]], dim=2)
+        v = rearrange(v, "b l d c -> (b l) d c")
+
+
+        b, _, _ = q.shape  ##actually bt
+        q, k, v = map(
+            lambda t: t.unsqueeze(3)
+            .reshape(b, t.shape[1], self.heads, self.dim_head)
+            .permute(0, 2, 1, 3)
+            .reshape(b * self.heads, t.shape[1], self.dim_head)
+            .contiguous(),
+            (q, k, v),
+        )
+
+        # actually compute the attention, what we cannot get enough of
+        if version.parse(xformers.__version__) >= version.parse("0.0.21"):
+            # NOTE: workaround for
+            # https://github.com/facebookresearch/xformers/issues/845
+            max_bs = 32768
+            N = q.shape[0]
+            n_batches = math.ceil(N / max_bs)
+            out = list()
+            for i_batch in range(n_batches):
+                batch = slice(i_batch * max_bs, (i_batch + 1) * max_bs)
+                out.append(
+                    xformers.ops.memory_efficient_attention(
+                        q[batch],
+                        k[batch],
+                        v[batch],
+                        attn_bias=None,
+                        op=self.attention_op,
+                    )
+                )
+            out = torch.cat(out, 0)
+        else:
+            out = xformers.ops.memory_efficient_attention(
+                q, k, v, attn_bias=None, op=self.attention_op
+            )
+
+        # TODO: Use this directly in the attention operation, as a bias
+        if exists(mask):
+            raise NotImplementedError
+        out = (
+            out.unsqueeze(0)
+            .reshape(b, self.heads, out.shape[1], self.dim_head)
+            .permute(0, 2, 1, 3)
+            .reshape(b, out.shape[1], self.heads * self.dim_head)
+        )
+        out = self.to_out(out)
+        out = rearrange(out, "bt (h w) c -> bt c h w", h=h, w=w, c=c)
+        return x + out 
+
+class Combiner(nn.Module):
+    def __init__(self, ch) -> None:
+        super().__init__()
+        self.conv = nn.Conv2d(ch,ch,1,padding=0)
+
+        nn.init.zeros_(self.conv.weight)
+        nn.init.zeros_(self.conv.bias)
+
+    def forward(self, x, context):
+        if self.training:
+            return checkpoint(self._forward, x, context, use_reentrant=False)
+        else:
+            return self._forward(x, context)
+    
+    def _forward(self, x, context):
+        ## x: b c h w, context: b c 2 h w
+        b, c, l, h, w = context.shape
+        bt, c, h, w = x.shape
+        context = rearrange(context, "b c l h w -> (b l) c h w")
+        context = self.conv(context)
+        context = rearrange(context, "(b l) c h w -> b c l h w", l=l)
+        x = rearrange(x, "(b t) c h w -> b c t h w", t=bt//b)
+        x[:,:,0] = x[:,:,0] + context[:,:,0]
+        x[:,:,-1] = x[:,:,-1] + context[:,:,1]
+        x = rearrange(x, "b c t h w -> (b t) c h w")
+        return x
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self,
+        *,
+        ch,
+        out_ch,
+        ch_mult=(1, 2, 4, 8),
+        num_res_blocks,
+        attn_resolutions,
+        dropout=0.0,
+        resamp_with_conv=True,
+        in_channels,
+        resolution,
+        z_channels,
+        give_pre_end=False,
+        tanh_out=False,
+        use_linear_attn=False,
+        attn_type="vanilla-xformers",
+        attn_level=[2,3], 
+        **ignorekwargs,
+    ):
+        super().__init__()
+        if use_linear_attn:
+            attn_type = "linear"
+        self.ch = ch
+        self.temb_ch = 0
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        self.resolution = resolution
+        self.in_channels = in_channels
+        self.give_pre_end = give_pre_end
+        self.tanh_out = tanh_out
+        self.attn_level = attn_level
+        # compute in_ch_mult, block_in and curr_res at lowest res
+        in_ch_mult = (1,) + tuple(ch_mult)
+        block_in = ch * ch_mult[self.num_resolutions - 1]
+        curr_res = resolution // 2 ** (self.num_resolutions - 1)
+        self.z_shape = (1, z_channels, curr_res, curr_res)
+        logpy.info(
+            "Working with z of shape {} = {} dimensions.".format(
+                self.z_shape, np.prod(self.z_shape)
+            )
+        )
+
+        make_attn_cls = self._make_attn()
+        make_resblock_cls = self._make_resblock()
+        make_conv_cls = self._make_conv()
+        # z to block_in
+        self.conv_in = torch.nn.Conv2d(
+            z_channels, block_in, kernel_size=3, stride=1, padding=1
+        )
+
+        # middle
+        self.mid = nn.Module()
+        self.mid.block_1 = make_resblock_cls(
+            in_channels=block_in,
+            out_channels=block_in,
+            temb_channels=self.temb_ch,
+            dropout=dropout,
+        )
+        self.mid.attn_1 = make_attn_cls(block_in, attn_type=attn_type)
+        self.mid.block_2 = make_resblock_cls(
+            in_channels=block_in,
+            out_channels=block_in,
+            temb_channels=self.temb_ch,
+            dropout=dropout,
+        )
+
+        # upsampling
+        self.up = nn.ModuleList()
+        self.attn_refinement = nn.ModuleList()
+        for i_level in reversed(range(self.num_resolutions)):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_out = ch * ch_mult[i_level]
+            for i_block in range(self.num_res_blocks + 1):
+                block.append(
+                    make_resblock_cls(
+                        in_channels=block_in,
+                        out_channels=block_out,
+                        temb_channels=self.temb_ch,
+                        dropout=dropout,
+                    )
+                )
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(make_attn_cls(block_in, attn_type=attn_type))
+            up = nn.Module()
+            up.block = block
+            up.attn = attn
+            if i_level != 0:
+                up.upsample = Upsample(block_in, resamp_with_conv)
+                curr_res = curr_res * 2
+            self.up.insert(0, up)  # prepend to get consistent order
+
+            if i_level in self.attn_level:
+                self.attn_refinement.insert(0, make_attn_cls(block_in, attn_type='memory-efficient-cross-attn-fusion', attn_kwargs={}))
+            else:
+                self.attn_refinement.insert(0, Combiner(block_in))
+        # end
+        self.norm_out = Normalize(block_in)
+        self.attn_refinement.append(Combiner(block_in))
+        self.conv_out = make_conv_cls(
+            block_in, out_ch, kernel_size=3, stride=1, padding=1
+        )
+
+    def _make_attn(self) -> Callable:
+        return make_attn
+
+    def _make_resblock(self) -> Callable:
+        return ResnetBlock
+
+    def _make_conv(self) -> Callable:
+        return torch.nn.Conv2d
+
+    def get_last_layer(self, **kwargs):
+        return self.conv_out.weight
+
+    def forward(self, z, ref_context=None, **kwargs):
+        ## ref_context: b c 2 h w, 2 means starting and ending frame
+        # assert z.shape[1:] == self.z_shape[1:]
+        self.last_z_shape = z.shape
+        # timestep embedding
+        temb = None
+
+        # z to block_in
+        h = self.conv_in(z)
+
+        # middle
+        h = self.mid.block_1(h, temb, **kwargs)
+        h = self.mid.attn_1(h, **kwargs)
+        h = self.mid.block_2(h, temb, **kwargs)
+
+        # upsampling
+        for i_level in reversed(range(self.num_resolutions)):
+            for i_block in range(self.num_res_blocks + 1):
+                h = self.up[i_level].block[i_block](h, temb, **kwargs)
+                if len(self.up[i_level].attn) > 0:
+                    h = self.up[i_level].attn[i_block](h, **kwargs)
+            if ref_context:
+                h = self.attn_refinement[i_level](x=h, context=ref_context[i_level])
+            if i_level != 0:
+                h = self.up[i_level].upsample(h)
+
+        # end
+        if self.give_pre_end:
+            return h
+
+        h = self.norm_out(h)
+        h = nonlinearity(h)
+        if ref_context:
+            # print(h.shape, ref_context[i_level].shape) #torch.Size([8, 128, 256, 256]) torch.Size([1, 128, 2, 256, 256])
+            h = self.attn_refinement[-1](x=h, context=ref_context[-1])
+        h = self.conv_out(h, **kwargs)
+        if self.tanh_out:
+            h = torch.tanh(h)
+        return h
+
+#####
+
+
+from abc import abstractmethod
+from lvdm.models.utils_diffusion import timestep_embedding
+
+from torch.utils.checkpoint import checkpoint
+from lvdm.basics import (
+    zero_module,
+    conv_nd,
+    linear,
+    normalization,
+)
+from lvdm.modules.networks.openaimodel3d import Upsample, Downsample
+class TimestepBlock(nn.Module):
+    """
+    Any module where forward() takes timestep embeddings as a second argument.
+    """
+
+    @abstractmethod
+    def forward(self, x: torch.Tensor, emb: torch.Tensor):
+        """
+        Apply the module to `x` given `emb` timestep embeddings.
+        """
+
+class ResBlock(TimestepBlock):
+    """
+    A residual block that can optionally change the number of channels.
+    :param channels: the number of input channels.
+    :param emb_channels: the number of timestep embedding channels.
+    :param dropout: the rate of dropout.
+    :param out_channels: if specified, the number of out channels.
+    :param use_conv: if True and out_channels is specified, use a spatial
+        convolution instead of a smaller 1x1 convolution to change the
+        channels in the skip connection.
+    :param dims: determines if the signal is 1D, 2D, or 3D.
+    :param use_checkpoint: if True, use gradient checkpointing on this module.
+    :param up: if True, use this block for upsampling.
+    :param down: if True, use this block for downsampling.
+    """
+
+    def __init__(
+        self,
+        channels: int,
+        emb_channels: int,
+        dropout: float,
+        out_channels: Optional[int] = None,
+        use_conv: bool = False,
+        use_scale_shift_norm: bool = False,
+        dims: int = 2,
+        use_checkpoint: bool = False,
+        up: bool = False,
+        down: bool = False,
+        kernel_size: int = 3,
+        exchange_temb_dims: bool = False,
+        skip_t_emb: bool = False,
+    ):
+        super().__init__()
+        self.channels = channels
+        self.emb_channels = emb_channels
+        self.dropout = dropout
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.use_checkpoint = use_checkpoint
+        self.use_scale_shift_norm = use_scale_shift_norm
+        self.exchange_temb_dims = exchange_temb_dims
+
+        if isinstance(kernel_size, Iterable):
+            padding = [k // 2 for k in kernel_size]
+        else:
+            padding = kernel_size // 2
+
+        self.in_layers = nn.Sequential(
+            normalization(channels),
+            nn.SiLU(),
+            conv_nd(dims, channels, self.out_channels, kernel_size, padding=padding),
+        )
+
+        self.updown = up or down
+
+        if up:
+            self.h_upd = Upsample(channels, False, dims)
+            self.x_upd = Upsample(channels, False, dims)
+        elif down:
+            self.h_upd = Downsample(channels, False, dims)
+            self.x_upd = Downsample(channels, False, dims)
+        else:
+            self.h_upd = self.x_upd = nn.Identity()
+
+        self.skip_t_emb = skip_t_emb
+        self.emb_out_channels = (
+            2 * self.out_channels if use_scale_shift_norm else self.out_channels
+        )
+        if self.skip_t_emb:
+            # print(f"Skipping timestep embedding in {self.__class__.__name__}")
+            assert not self.use_scale_shift_norm
+            self.emb_layers = None
+            self.exchange_temb_dims = False
+        else:
+            self.emb_layers = nn.Sequential(
+                nn.SiLU(),
+                linear(
+                    emb_channels,
+                    self.emb_out_channels,
+                ),
+            )
+
+        self.out_layers = nn.Sequential(
+            normalization(self.out_channels),
+            nn.SiLU(),
+            nn.Dropout(p=dropout),
+            zero_module(
+                conv_nd(
+                    dims,
+                    self.out_channels,
+                    self.out_channels,
+                    kernel_size,
+                    padding=padding,
+                )
+            ),
+        )
+
+        if self.out_channels == channels:
+            self.skip_connection = nn.Identity()
+        elif use_conv:
+            self.skip_connection = conv_nd(
+                dims, channels, self.out_channels, kernel_size, padding=padding
+            )
+        else:
+            self.skip_connection = conv_nd(dims, channels, self.out_channels, 1)
+
+    def forward(self, x: torch.Tensor, emb: torch.Tensor) -> torch.Tensor:
+        """
+        Apply the block to a Tensor, conditioned on a timestep embedding.
+        :param x: an [N x C x ...] Tensor of features.
+        :param emb: an [N x emb_channels] Tensor of timestep embeddings.
+        :return: an [N x C x ...] Tensor of outputs.
+        """
+        if self.use_checkpoint:
+            return checkpoint(self._forward, x, emb, use_reentrant=False)
+        else:
+            return self._forward(x, emb)
+
+    def _forward(self, x: torch.Tensor, emb: torch.Tensor) -> torch.Tensor:
+        if self.updown:
+            in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]
+            h = in_rest(x)
+            h = self.h_upd(h)
+            x = self.x_upd(x)
+            h = in_conv(h)
+        else:
+            h = self.in_layers(x)
+
+        if self.skip_t_emb:
+            emb_out = torch.zeros_like(h)
+        else:
+            emb_out = self.emb_layers(emb).type(h.dtype)
+        while len(emb_out.shape) < len(h.shape):
+            emb_out = emb_out[..., None]
+        if self.use_scale_shift_norm:
+            out_norm, out_rest = self.out_layers[0], self.out_layers[1:]
+            scale, shift = torch.chunk(emb_out, 2, dim=1)
+            h = out_norm(h) * (1 + scale) + shift
+            h = out_rest(h)
+        else:
+            if self.exchange_temb_dims:
+                emb_out = rearrange(emb_out, "b t c ... -> b c t ...")
+            h = h + emb_out
+            h = self.out_layers(h)
+        return self.skip_connection(x) + h
+#####
+
+#####
+from lvdm.modules.attention_svd import *
+class VideoTransformerBlock(nn.Module):
+    ATTENTION_MODES = {
+        "softmax": CrossAttention,
+        "softmax-xformers": MemoryEfficientCrossAttention,
+    }
+
+    def __init__(
+        self,
+        dim,
+        n_heads,
+        d_head,
+        dropout=0.0,
+        context_dim=None,
+        gated_ff=True,
+        checkpoint=True,
+        timesteps=None,
+        ff_in=False,
+        inner_dim=None,
+        attn_mode="softmax",
+        disable_self_attn=False,
+        disable_temporal_crossattention=False,
+        switch_temporal_ca_to_sa=False,
+    ):
+        super().__init__()
+
+        attn_cls = self.ATTENTION_MODES[attn_mode]
+
+        self.ff_in = ff_in or inner_dim is not None
+        if inner_dim is None:
+            inner_dim = dim
+
+        assert int(n_heads * d_head) == inner_dim
+
+        self.is_res = inner_dim == dim
+
+        if self.ff_in:
+            self.norm_in = nn.LayerNorm(dim)
+            self.ff_in = FeedForward(
+                dim, dim_out=inner_dim, dropout=dropout, glu=gated_ff
+            )
+
+        self.timesteps = timesteps
+        self.disable_self_attn = disable_self_attn
+        if self.disable_self_attn:
+            self.attn1 = attn_cls(
+                query_dim=inner_dim,
+                heads=n_heads,
+                dim_head=d_head,
+                context_dim=context_dim,
+                dropout=dropout,
+            )  # is a cross-attention
+        else:
+            self.attn1 = attn_cls(
+                query_dim=inner_dim, heads=n_heads, dim_head=d_head, dropout=dropout
+            )  # is a self-attention
+
+        self.ff = FeedForward(inner_dim, dim_out=dim, dropout=dropout, glu=gated_ff)
+
+        if disable_temporal_crossattention:
+            if switch_temporal_ca_to_sa:
+                raise ValueError
+            else:
+                self.attn2 = None
+        else:
+            self.norm2 = nn.LayerNorm(inner_dim)
+            if switch_temporal_ca_to_sa:
+                self.attn2 = attn_cls(
+                    query_dim=inner_dim, heads=n_heads, dim_head=d_head, dropout=dropout
+                )  # is a self-attention
+            else:
+                self.attn2 = attn_cls(
+                    query_dim=inner_dim,
+                    context_dim=context_dim,
+                    heads=n_heads,
+                    dim_head=d_head,
+                    dropout=dropout,
+                )  # is self-attn if context is none
+
+        self.norm1 = nn.LayerNorm(inner_dim)
+        self.norm3 = nn.LayerNorm(inner_dim)
+        self.switch_temporal_ca_to_sa = switch_temporal_ca_to_sa
+
+        self.checkpoint = checkpoint
+        if self.checkpoint:
+            print(f"====>{self.__class__.__name__} is using checkpointing")
+        else:
+            print(f"====>{self.__class__.__name__} is NOT using checkpointing")
+
+    def forward(
+        self, x: torch.Tensor, context: torch.Tensor = None, timesteps: int = None
+    ) -> torch.Tensor:
+        if self.checkpoint:
+            return checkpoint(self._forward, x, context, timesteps, use_reentrant=False)
+        else:
+            return self._forward(x, context, timesteps=timesteps)
+
+    def _forward(self, x, context=None, timesteps=None):
+        assert self.timesteps or timesteps
+        assert not (self.timesteps and timesteps) or self.timesteps == timesteps
+        timesteps = self.timesteps or timesteps
+        B, S, C = x.shape
+        x = rearrange(x, "(b t) s c -> (b s) t c", t=timesteps)
+
+        if self.ff_in:
+            x_skip = x
+            x = self.ff_in(self.norm_in(x))
+            if self.is_res:
+                x += x_skip
+
+        if self.disable_self_attn:
+            x = self.attn1(self.norm1(x), context=context) + x
+        else:
+            x = self.attn1(self.norm1(x)) + x
+
+        if self.attn2 is not None:
+            if self.switch_temporal_ca_to_sa:
+                x = self.attn2(self.norm2(x)) + x
+            else:
+                x = self.attn2(self.norm2(x), context=context) + x
+        x_skip = x
+        x = self.ff(self.norm3(x))
+        if self.is_res:
+            x += x_skip
+
+        x = rearrange(
+            x, "(b s) t c -> (b t) s c", s=S, b=B // timesteps, c=C, t=timesteps
+        )
+        return x
+
+    def get_last_layer(self):
+        return self.ff.net[-1].weight
+
+#####
+
+#####
+import functools
+def partialclass(cls, *args, **kwargs):
+    class NewCls(cls):
+        __init__ = functools.partialmethod(cls.__init__, *args, **kwargs)
+
+    return NewCls
+######
+
+class VideoResBlock(ResnetBlock):
+    def __init__(
+        self,
+        out_channels,
+        *args,
+        dropout=0.0,
+        video_kernel_size=3,
+        alpha=0.0,
+        merge_strategy="learned",
+        **kwargs,
+    ):
+        super().__init__(out_channels=out_channels, dropout=dropout, *args, **kwargs)
+        if video_kernel_size is None:
+            video_kernel_size = [3, 1, 1]
+        self.time_stack = ResBlock(
+            channels=out_channels,
+            emb_channels=0,
+            dropout=dropout,
+            dims=3,
+            use_scale_shift_norm=False,
+            use_conv=False,
+            up=False,
+            down=False,
+            kernel_size=video_kernel_size,
+            use_checkpoint=True,
+            skip_t_emb=True,
+        )
+
+        self.merge_strategy = merge_strategy
+        if self.merge_strategy == "fixed":
+            self.register_buffer("mix_factor", torch.Tensor([alpha]))
+        elif self.merge_strategy == "learned":
+            self.register_parameter(
+                "mix_factor", torch.nn.Parameter(torch.Tensor([alpha]))
+            )
+        else:
+            raise ValueError(f"unknown merge strategy {self.merge_strategy}")
+
+    def get_alpha(self, bs):
+        if self.merge_strategy == "fixed":
+            return self.mix_factor
+        elif self.merge_strategy == "learned":
+            return torch.sigmoid(self.mix_factor)
+        else:
+            raise NotImplementedError()
+
+    def forward(self, x, temb, skip_video=False, timesteps=None):
+        if timesteps is None:
+            timesteps = self.timesteps
+
+        b, c, h, w = x.shape
+
+        x = super().forward(x, temb)
+
+        if not skip_video:
+            x_mix = rearrange(x, "(b t) c h w -> b c t h w", t=timesteps)
+
+            x = rearrange(x, "(b t) c h w -> b c t h w", t=timesteps)
+
+            x = self.time_stack(x, temb)
+
+            alpha = self.get_alpha(bs=b // timesteps)
+            x = alpha * x + (1.0 - alpha) * x_mix
+
+            x = rearrange(x, "b c t h w -> (b t) c h w")
+        return x
+
+
+class AE3DConv(torch.nn.Conv2d):
+    def __init__(self, in_channels, out_channels, video_kernel_size=3, *args, **kwargs):
+        super().__init__(in_channels, out_channels, *args, **kwargs)
+        if isinstance(video_kernel_size, Iterable):
+            padding = [int(k // 2) for k in video_kernel_size]
+        else:
+            padding = int(video_kernel_size // 2)
+
+        self.time_mix_conv = torch.nn.Conv3d(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=video_kernel_size,
+            padding=padding,
+        )
+
+    def forward(self, input, timesteps, skip_video=False):
+        x = super().forward(input)
+        if skip_video:
+            return x
+        x = rearrange(x, "(b t) c h w -> b c t h w", t=timesteps)
+        x = self.time_mix_conv(x)
+        return rearrange(x, "b c t h w -> (b t) c h w")
+
+
+class VideoBlock(AttnBlock):
+    def __init__(
+        self, in_channels: int, alpha: float = 0, merge_strategy: str = "learned"
+    ):
+        super().__init__(in_channels)
+        # no context, single headed, as in base class
+        self.time_mix_block = VideoTransformerBlock(
+            dim=in_channels,
+            n_heads=1,
+            d_head=in_channels,
+            checkpoint=True,
+            ff_in=True,
+            attn_mode="softmax",
+        )
+
+        time_embed_dim = self.in_channels * 4
+        self.video_time_embed = torch.nn.Sequential(
+            torch.nn.Linear(self.in_channels, time_embed_dim),
+            torch.nn.SiLU(),
+            torch.nn.Linear(time_embed_dim, self.in_channels),
+        )
+
+        self.merge_strategy = merge_strategy
+        if self.merge_strategy == "fixed":
+            self.register_buffer("mix_factor", torch.Tensor([alpha]))
+        elif self.merge_strategy == "learned":
+            self.register_parameter(
+                "mix_factor", torch.nn.Parameter(torch.Tensor([alpha]))
+            )
+        else:
+            raise ValueError(f"unknown merge strategy {self.merge_strategy}")
+
+    def forward(self, x, timesteps, skip_video=False):
+        if skip_video:
+            return super().forward(x)
+
+        x_in = x
+        x = self.attention(x)
+        h, w = x.shape[2:]
+        x = rearrange(x, "b c h w -> b (h w) c")
+
+        x_mix = x
+        num_frames = torch.arange(timesteps, device=x.device)
+        num_frames = repeat(num_frames, "t -> b t", b=x.shape[0] // timesteps)
+        num_frames = rearrange(num_frames, "b t -> (b t)")
+        t_emb = timestep_embedding(num_frames, self.in_channels, repeat_only=False)
+        emb = self.video_time_embed(t_emb)  # b, n_channels
+        emb = emb[:, None, :]
+        x_mix = x_mix + emb
+
+        alpha = self.get_alpha()
+        x_mix = self.time_mix_block(x_mix, timesteps=timesteps)
+        x = alpha * x + (1.0 - alpha) * x_mix  # alpha merge
+
+        x = rearrange(x, "b (h w) c -> b c h w", h=h, w=w)
+        x = self.proj_out(x)
+
+        return x_in + x
+
+    def get_alpha(
+        self,
+    ):
+        if self.merge_strategy == "fixed":
+            return self.mix_factor
+        elif self.merge_strategy == "learned":
+            return torch.sigmoid(self.mix_factor)
+        else:
+            raise NotImplementedError(f"unknown merge strategy {self.merge_strategy}")
+
+
+class MemoryEfficientVideoBlock(MemoryEfficientAttnBlock):
+    def __init__(
+        self, in_channels: int, alpha: float = 0, merge_strategy: str = "learned"
+    ):
+        super().__init__(in_channels)
+        # no context, single headed, as in base class
+        self.time_mix_block = VideoTransformerBlock(
+            dim=in_channels,
+            n_heads=1,
+            d_head=in_channels,
+            checkpoint=True,
+            ff_in=True,
+            attn_mode="softmax-xformers",
+        )
+
+        time_embed_dim = self.in_channels * 4
+        self.video_time_embed = torch.nn.Sequential(
+            torch.nn.Linear(self.in_channels, time_embed_dim),
+            torch.nn.SiLU(),
+            torch.nn.Linear(time_embed_dim, self.in_channels),
+        )
+
+        self.merge_strategy = merge_strategy
+        if self.merge_strategy == "fixed":
+            self.register_buffer("mix_factor", torch.Tensor([alpha]))
+        elif self.merge_strategy == "learned":
+            self.register_parameter(
+                "mix_factor", torch.nn.Parameter(torch.Tensor([alpha]))
+            )
+        else:
+            raise ValueError(f"unknown merge strategy {self.merge_strategy}")
+
+    def forward(self, x, timesteps, skip_time_block=False):
+        if skip_time_block:
+            return super().forward(x)
+
+        x_in = x
+        x = self.attention(x)
+        h, w = x.shape[2:]
+        x = rearrange(x, "b c h w -> b (h w) c")
+
+        x_mix = x
+        num_frames = torch.arange(timesteps, device=x.device)
+        num_frames = repeat(num_frames, "t -> b t", b=x.shape[0] // timesteps)
+        num_frames = rearrange(num_frames, "b t -> (b t)")
+        t_emb = timestep_embedding(num_frames, self.in_channels, repeat_only=False)
+        emb = self.video_time_embed(t_emb)  # b, n_channels
+        emb = emb[:, None, :]
+        x_mix = x_mix + emb
+
+        alpha = self.get_alpha()
+        x_mix = self.time_mix_block(x_mix, timesteps=timesteps)
+        x = alpha * x + (1.0 - alpha) * x_mix  # alpha merge
+
+        x = rearrange(x, "b (h w) c -> b c h w", h=h, w=w)
+        x = self.proj_out(x)
+
+        return x_in + x
+
+    def get_alpha(
+        self,
+    ):
+        if self.merge_strategy == "fixed":
+            return self.mix_factor
+        elif self.merge_strategy == "learned":
+            return torch.sigmoid(self.mix_factor)
+        else:
+            raise NotImplementedError(f"unknown merge strategy {self.merge_strategy}")
+
+
+def make_time_attn(
+    in_channels,
+    attn_type="vanilla",
+    attn_kwargs=None,
+    alpha: float = 0,
+    merge_strategy: str = "learned",
+):
+    assert attn_type in [
+        "vanilla",
+        "vanilla-xformers",
+    ], f"attn_type {attn_type} not supported for spatio-temporal attention"
+    print(
+        f"making spatial and temporal attention of type '{attn_type}' with {in_channels} in_channels"
+    )
+    if not XFORMERS_IS_AVAILABLE and attn_type == "vanilla-xformers":
+        print(
+            f"Attention mode '{attn_type}' is not available. Falling back to vanilla attention. "
+            f"This is not a problem in Pytorch >= 2.0. FYI, you are running with PyTorch version {torch.__version__}"
+        )
+        attn_type = "vanilla"
+
+    if attn_type == "vanilla":
+        assert attn_kwargs is None
+        return partialclass(
+            VideoBlock, in_channels, alpha=alpha, merge_strategy=merge_strategy
+        )
+    elif attn_type == "vanilla-xformers":
+        print(f"building MemoryEfficientAttnBlock with {in_channels} in_channels...")
+        return partialclass(
+            MemoryEfficientVideoBlock,
+            in_channels,
+            alpha=alpha,
+            merge_strategy=merge_strategy,
+        )
+    else:
+        return NotImplementedError()
+
+
+class Conv2DWrapper(torch.nn.Conv2d):
+    def forward(self, input: torch.Tensor, **kwargs) -> torch.Tensor:
+        return super().forward(input)
+
+
+class VideoDecoder(Decoder):
+    available_time_modes = ["all", "conv-only", "attn-only"]
+
+    def __init__(
+        self,
+        *args,
+        video_kernel_size: Union[int, list] = [3,1,1],
+        alpha: float = 0.0,
+        merge_strategy: str = "learned",
+        time_mode: str = "conv-only",
+        **kwargs,
+    ):
+        self.video_kernel_size = video_kernel_size
+        self.alpha = alpha
+        self.merge_strategy = merge_strategy
+        self.time_mode = time_mode
+        assert (
+            self.time_mode in self.available_time_modes
+        ), f"time_mode parameter has to be in {self.available_time_modes}"
+        super().__init__(*args, **kwargs)
+
+    def get_last_layer(self, skip_time_mix=False, **kwargs):
+        if self.time_mode == "attn-only":
+            raise NotImplementedError("TODO")
+        else:
+            return (
+                self.conv_out.time_mix_conv.weight
+                if not skip_time_mix
+                else self.conv_out.weight
+            )
+
+    def _make_attn(self) -> Callable:
+        if self.time_mode not in ["conv-only", "only-last-conv"]:
+            return partialclass(
+                make_time_attn,
+                alpha=self.alpha,
+                merge_strategy=self.merge_strategy,
+            )
+        else:
+            return super()._make_attn()
+
+    def _make_conv(self) -> Callable:
+        if self.time_mode != "attn-only":
+            return partialclass(AE3DConv, video_kernel_size=self.video_kernel_size)
+        else:
+            return Conv2DWrapper
+
+    def _make_resblock(self) -> Callable:
+        if self.time_mode not in ["attn-only", "only-last-conv"]:
+            return partialclass(
+                VideoResBlock,
+                video_kernel_size=self.video_kernel_size,
+                alpha=self.alpha,
+                merge_strategy=self.merge_strategy,
+            )
+        else:
+            return super()._make_resblock()