Upload 106 files

.gitattributes

@@ -33,3 +33,6 @@ 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
+examples/kaifu_resize.png filter=lfs diff=lfs merge=lfs -text
+examples/sam_resize.png filter=lfs diff=lfs merge=lfs -text
+examples/schmidhuber_resize.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,3 +1,14 @@
----
-license: apache-2.0
----
+---
+title: InstantID
+emoji: 😻
+colorFrom: gray
+colorTo: gray
+sdk: gradio
+sdk_version: 4.15.0
+app_file: app.py
+pinned: false
+license: apache-2.0
+disable_embedding: true
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,653 @@
+import cv2
+import torch
+import random
+import numpy as np
+
+import spaces
+
+import PIL
+from PIL import Image
+from typing import Tuple
+
+import diffusers
+from diffusers.utils import load_image
+from diffusers.models import ControlNetModel
+from diffusers.pipelines.controlnet.multicontrolnet import MultiControlNetModel
+
+from huggingface_hub import hf_hub_download
+
+from insightface.app import FaceAnalysis
+
+from style_template import styles
+from pipeline_stable_diffusion_xl_instantid_full import StableDiffusionXLInstantIDPipeline, draw_kps
+
+from controlnet_aux import OpenposeDetector
+
+import gradio as gr
+
+from depth_anything.dpt import DepthAnything
+from depth_anything.util.transform import Resize, NormalizeImage, PrepareForNet
+
+import torch.nn.functional as F
+from torchvision.transforms import Compose
+
+# global variable
+MAX_SEED = np.iinfo(np.int32).max
+device = "cuda" if torch.cuda.is_available() else "cpu"
+dtype = torch.float16 if str(device).__contains__("cuda") else torch.float32
+STYLE_NAMES = list(styles.keys())
+DEFAULT_STYLE_NAME = "Spring Festival"
+enable_lcm_arg = False
+
+# download checkpoints
+from huggingface_hub import hf_hub_download
+
+hf_hub_download(repo_id="InstantX/InstantID", filename="ControlNetModel/config.json", local_dir="./checkpoints")
+hf_hub_download(
+    repo_id="InstantX/InstantID",
+    filename="ControlNetModel/diffusion_pytorch_model.safetensors",
+    local_dir="./checkpoints",
+)
+hf_hub_download(repo_id="InstantX/InstantID", filename="ip-adapter.bin", local_dir="./checkpoints")
+
+# Load face encoder
+app = FaceAnalysis(
+    name="antelopev2",
+    root="./",
+    providers=["CPUExecutionProvider"],
+)
+app.prepare(ctx_id=0, det_size=(640, 640))
+
+openpose = OpenposeDetector.from_pretrained("lllyasviel/ControlNet")
+
+depth_anything = DepthAnything.from_pretrained('LiheYoung/depth_anything_vitl14').to(device).eval()
+
+transform = Compose([
+    Resize(
+        width=518,
+        height=518,
+        resize_target=False,
+        keep_aspect_ratio=True,
+        ensure_multiple_of=14,
+        resize_method='lower_bound',
+        image_interpolation_method=cv2.INTER_CUBIC,
+    ),
+    NormalizeImage(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
+    PrepareForNet(),
+])
+
+# Path to InstantID models
+face_adapter = f"./checkpoints/ip-adapter.bin"
+controlnet_path = f"./checkpoints/ControlNetModel"
+
+# Load pipeline face ControlNetModel
+controlnet_identitynet = ControlNetModel.from_pretrained(
+    controlnet_path, torch_dtype=dtype
+)
+
+# controlnet-pose/canny/depth
+controlnet_pose_model = "thibaud/controlnet-openpose-sdxl-1.0"
+controlnet_canny_model = "diffusers/controlnet-canny-sdxl-1.0"
+controlnet_depth_model = "diffusers/controlnet-depth-sdxl-1.0-small"
+
+controlnet_pose = ControlNetModel.from_pretrained(
+    controlnet_pose_model, torch_dtype=dtype
+).to(device)
+controlnet_canny = ControlNetModel.from_pretrained(
+    controlnet_canny_model, torch_dtype=dtype
+).to(device)
+controlnet_depth = ControlNetModel.from_pretrained(
+    controlnet_depth_model, torch_dtype=dtype
+).to(device)
+
+def get_depth_map(image):
+    
+    image = np.array(image) / 255.0
+
+    h, w = image.shape[:2]
+
+    image = transform({'image': image})['image']
+    image = torch.from_numpy(image).unsqueeze(0).to("cuda")
+
+    with torch.no_grad():
+        depth = depth_anything(image)
+
+    depth = F.interpolate(depth[None], (h, w), mode='bilinear', align_corners=False)[0, 0]
+    depth = (depth - depth.min()) / (depth.max() - depth.min()) * 255.0
+
+    depth = depth.cpu().numpy().astype(np.uint8)
+
+    depth_image = Image.fromarray(depth)
+
+    return depth_image
+
+def get_canny_image(image, t1=100, t2=200):
+    image = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
+    edges = cv2.Canny(image, t1, t2)
+    return Image.fromarray(edges, "L")
+
+controlnet_map = {
+    "pose": controlnet_pose,
+    "canny": controlnet_canny,
+    "depth": controlnet_depth,
+}
+controlnet_map_fn = {
+    "pose": openpose,
+    "canny": get_canny_image,
+    "depth": get_depth_map,
+}
+
+pretrained_model_name_or_path = "wangqixun/YamerMIX_v8"
+
+pipe = StableDiffusionXLInstantIDPipeline.from_pretrained(
+    pretrained_model_name_or_path,
+    controlnet=[controlnet_identitynet],
+    torch_dtype=dtype,
+    safety_checker=None,
+    feature_extractor=None,
+).to(device)
+
+pipe.scheduler = diffusers.EulerDiscreteScheduler.from_config(
+    pipe.scheduler.config
+)
+
+# load and disable LCM
+pipe.load_lora_weights("latent-consistency/lcm-lora-sdxl")
+pipe.disable_lora()
+
+pipe.cuda()
+pipe.load_ip_adapter_instantid(face_adapter)
+pipe.image_proj_model.to("cuda")
+pipe.unet.to("cuda")
+
+def toggle_lcm_ui(value):
+    if value:
+        return (
+            gr.update(minimum=0, maximum=100, step=1, value=5),
+            gr.update(minimum=0.1, maximum=20.0, step=0.1, value=1.5),
+        )
+    else:
+        return (
+            gr.update(minimum=5, maximum=100, step=1, value=30),
+            gr.update(minimum=0.1, maximum=20.0, step=0.1, value=5),
+        )
+
+def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
+    if randomize_seed:
+        seed = random.randint(0, MAX_SEED)
+    return seed
+
+def remove_tips():
+    return gr.update(visible=False)
+
+def get_example():
+    case = [
+        [
+            "./examples/yann-lecun_resize.jpg",
+            None,
+            "a man",
+            "Spring Festival",
+            "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
+        ],
+        [
+            "./examples/musk_resize.jpeg",
+            "./examples/poses/pose2.jpg",
+            "a man flying in the sky in Mars",
+            "Mars",
+            "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
+        ],
+        [
+            "./examples/sam_resize.png",
+            "./examples/poses/pose4.jpg",
+            "a man doing a silly pose wearing a suite",
+            "Jungle",
+            "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, gree",
+        ],
+        [
+            "./examples/schmidhuber_resize.png",
+            "./examples/poses/pose3.jpg",
+            "a man sit on a chair",
+            "Neon",
+            "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
+        ],
+        [
+            "./examples/kaifu_resize.png",
+            "./examples/poses/pose.jpg",
+            "a man",
+            "Vibrant Color",
+            "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
+        ],
+    ]
+    return case
+
+def run_for_examples(face_file, pose_file, prompt, style, negative_prompt):
+    return generate_image(
+        face_file,
+        pose_file,
+        prompt,
+        negative_prompt,
+        style,
+        20,  # num_steps
+        0.8,  # identitynet_strength_ratio
+        0.8,  # adapter_strength_ratio
+        0.4,  # pose_strength
+        0.3,  # canny_strength
+        0.5,  # depth_strength
+        ["pose", "canny"],  # controlnet_selection
+        5.0,  # guidance_scale
+        42,  # seed
+        "EulerDiscreteScheduler",  # scheduler
+        False,  # enable_LCM
+        True,  # enable_Face_Region
+    )
+
+def convert_from_cv2_to_image(img: np.ndarray) -> Image:
+    return Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
+
+def convert_from_image_to_cv2(img: Image) -> np.ndarray:
+    return cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
+
+def resize_img(
+    input_image,
+    max_side=1280,
+    min_side=1024,
+    size=None,
+    pad_to_max_side=False,
+    mode=PIL.Image.BILINEAR,
+    base_pixel_number=64,
+):
+    w, h = input_image.size
+    if size is not None:
+        w_resize_new, h_resize_new = size
+    else:
+        ratio = min_side / min(h, w)
+        w, h = round(ratio * w), round(ratio * h)
+        ratio = max_side / max(h, w)
+        input_image = input_image.resize([round(ratio * w), round(ratio * h)], mode)
+        w_resize_new = (round(ratio * w) // base_pixel_number) * base_pixel_number
+        h_resize_new = (round(ratio * h) // base_pixel_number) * base_pixel_number
+    input_image = input_image.resize([w_resize_new, h_resize_new], mode)
+
+    if pad_to_max_side:
+        res = np.ones([max_side, max_side, 3], dtype=np.uint8) * 255
+        offset_x = (max_side - w_resize_new) // 2
+        offset_y = (max_side - h_resize_new) // 2
+        res[
+            offset_y : offset_y + h_resize_new, offset_x : offset_x + w_resize_new
+        ] = np.array(input_image)
+        input_image = Image.fromarray(res)
+    return input_image
+
+def apply_style(
+    style_name: str, positive: str, negative: str = ""
+) -> Tuple[str, str]:
+    p, n = styles.get(style_name, styles[DEFAULT_STYLE_NAME])
+    return p.replace("{prompt}", positive), n + " " + negative
+
+@spaces.GPU
+def generate_image(
+    face_image_path,
+    pose_image_path,
+    prompt,
+    negative_prompt,
+    style_name,
+    num_steps,
+    identitynet_strength_ratio,
+    adapter_strength_ratio,
+    pose_strength,
+    canny_strength,
+    depth_strength,
+    controlnet_selection,
+    guidance_scale,
+    seed,
+    scheduler,
+    enable_LCM,
+    enhance_face_region,
+    progress=gr.Progress(track_tqdm=True),
+):
+
+    if enable_LCM:
+        pipe.scheduler = diffusers.LCMScheduler.from_config(pipe.scheduler.config)
+        pipe.enable_lora()
+    else:
+        pipe.disable_lora()
+        scheduler_class_name = scheduler.split("-")[0]
+
+        add_kwargs = {}
+        if len(scheduler.split("-")) > 1:
+            add_kwargs["use_karras_sigmas"] = True
+        if len(scheduler.split("-")) > 2:
+            add_kwargs["algorithm_type"] = "sde-dpmsolver++"
+        scheduler = getattr(diffusers, scheduler_class_name)
+        pipe.scheduler = scheduler.from_config(pipe.scheduler.config, **add_kwargs)
+
+    if face_image_path is None:
+        raise gr.Error(
+            f"Cannot find any input face image! Please upload the face image"
+        )
+
+    if prompt is None:
+        prompt = "a person"
+
+    # apply the style template
+    prompt, negative_prompt = apply_style(style_name, prompt, negative_prompt)
+
+    face_image = load_image(face_image_path)
+    face_image = resize_img(face_image, max_side=1024)
+    face_image_cv2 = convert_from_image_to_cv2(face_image)
+    height, width, _ = face_image_cv2.shape
+
+    # Extract face features
+    face_info = app.get(face_image_cv2)
+
+    if len(face_info) == 0:
+        raise gr.Error(
+            f"Unable to detect a face in the image. Please upload a different photo with a clear face."
+        )
+
+    face_info = sorted(
+        face_info,
+        key=lambda x: (x["bbox"][2] - x["bbox"][0]) * x["bbox"][3] - x["bbox"][1],
+    )[
+        -1
+    ]  # only use the maximum face
+    face_emb = face_info["embedding"]
+    face_kps = draw_kps(convert_from_cv2_to_image(face_image_cv2), face_info["kps"])
+    img_controlnet = face_image
+    if pose_image_path is not None:
+        pose_image = load_image(pose_image_path)
+        pose_image = resize_img(pose_image, max_side=1024)
+        img_controlnet = pose_image
+        pose_image_cv2 = convert_from_image_to_cv2(pose_image)
+
+        face_info = app.get(pose_image_cv2)
+
+        if len(face_info) == 0:
+            raise gr.Error(
+                f"Cannot find any face in the reference image! Please upload another person image"
+            )
+
+        face_info = face_info[-1]
+        face_kps = draw_kps(pose_image, face_info["kps"])
+
+        width, height = face_kps.size
+
+    if enhance_face_region:
+        control_mask = np.zeros([height, width, 3])
+        x1, y1, x2, y2 = face_info["bbox"]
+        x1, y1, x2, y2 = int(x1), int(y1), int(x2), int(y2)
+        control_mask[y1:y2, x1:x2] = 255
+        control_mask = Image.fromarray(control_mask.astype(np.uint8))
+    else:
+        control_mask = None
+
+    if len(controlnet_selection) > 0:
+        controlnet_scales = {
+            "pose": pose_strength,
+            "canny": canny_strength,
+            "depth": depth_strength,
+        }
+        pipe.controlnet = MultiControlNetModel(
+            [controlnet_identitynet]
+            + [controlnet_map[s] for s in controlnet_selection]
+        )
+        control_scales = [float(identitynet_strength_ratio)] + [
+            controlnet_scales[s] for s in controlnet_selection
+        ]
+        control_images = [face_kps] + [
+            controlnet_map_fn[s](img_controlnet).resize((width, height))
+            for s in controlnet_selection
+        ]
+    else:
+        pipe.controlnet = controlnet_identitynet
+        control_scales = float(identitynet_strength_ratio)
+        control_images = face_kps
+
+    generator = torch.Generator(device=device).manual_seed(seed)
+
+    print("Start inference...")
+    print(f"[Debug] Prompt: {prompt}, \n[Debug] Neg Prompt: {negative_prompt}")
+
+    pipe.set_ip_adapter_scale(adapter_strength_ratio)
+    images = pipe(
+        prompt=prompt,
+        negative_prompt=negative_prompt,
+        image_embeds=face_emb,
+        image=control_images,
+        control_mask=control_mask,
+        controlnet_conditioning_scale=control_scales,
+        num_inference_steps=num_steps,
+        guidance_scale=guidance_scale,
+        height=height,
+        width=width,
+        generator=generator,
+    ).images
+
+    return images[0], gr.update(visible=True)
+
+# Description
+title = r"""
+<h1 align="center">InstantID: Zero-shot Identity-Preserving Generation in Seconds</h1>
+"""
+
+description = r"""
+<b>Official 🤗 Gradio demo</b> for <a href='https://github.com/InstantID/InstantID' target='_blank'><b>InstantID: Zero-shot Identity-Preserving Generation in Seconds</b></a>.<br>
+
+We are organizing a Spring Festival event with HuggingFace from 2.7 to 2.25, and you can now generate pictures of Spring Festival costumes. Happy Dragon Year 🐲 ! Share the joy with your family.<br>
+
+How to use:<br>
+1. Upload an image with a face. For images with multiple faces, we will only detect the largest face. Ensure the face is not too small and is clearly visible without significant obstructions or blurring.
+2. (Optional) You can upload another image as a reference for the face pose. If you don't, we will use the first detected face image to extract facial landmarks. If you use a cropped face at step 1, it is recommended to upload it to define a new face pose.
+3. (Optional) You can select multiple ControlNet models to control the generation process. The default is to use the IdentityNet only. The ControlNet models include pose skeleton, canny, and depth. You can adjust the strength of each ControlNet model to control the generation process.
+4. Enter a text prompt, as done in normal text-to-image models.
+5. Click the <b>Submit</b> button to begin customization.
+6. Share your customized photo with your friends and enjoy! 😊"""
+
+article = r"""
+---
+📝 **Citation**
+<br>
+If our work is helpful for your research or applications, please cite us via:
+```bibtex
+@article{wang2024instantid,
+  title={InstantID: Zero-shot Identity-Preserving Generation in Seconds},
+  author={Wang, Qixun and Bai, Xu and Wang, Haofan and Qin, Zekui and Chen, Anthony},
+  journal={arXiv preprint arXiv:2401.07519},
+  year={2024}
+}
+```
+📧 **Contact**
+<br>
+If you have any questions, please feel free to open an issue or directly reach us out at <b>haofanwang.ai@gmail.com</b>.
+"""
+
+tips = r"""
+### Usage tips of InstantID
+1. If you're not satisfied with the similarity, try increasing the weight of "IdentityNet Strength" and "Adapter Strength."    
+2. If you feel that the saturation is too high, first decrease the Adapter strength. If it remains too high, then decrease the IdentityNet strength.
+3. If you find that text control is not as expected, decrease Adapter strength.
+4. If you find that realistic style is not good enough, go for our Github repo and use a more realistic base model.
+"""
+
+css = """
+.gradio-container {width: 85% !important}
+"""
+with gr.Blocks(css=css) as demo:
+    # description
+    gr.Markdown(title)
+    gr.Markdown(description)
+
+    with gr.Row():
+        with gr.Column():
+            with gr.Row(equal_height=True):
+                # upload face image
+                face_file = gr.Image(
+                    label="Upload a photo of your face", type="filepath"
+                )
+                # optional: upload a reference pose image
+                pose_file = gr.Image(
+                    label="Upload a reference pose image (Optional)",
+                    type="filepath",
+                )
+
+            # prompt
+            prompt = gr.Textbox(
+                label="Prompt",
+                info="Give simple prompt is enough to achieve good face fidelity",
+                placeholder="A photo of a person",
+                value="",
+            )
+
+            submit = gr.Button("Submit", variant="primary")
+            enable_LCM = gr.Checkbox(
+                label="Enable Fast Inference with LCM", value=enable_lcm_arg,
+                info="LCM speeds up the inference step, the trade-off is the quality of the generated image. It performs better with portrait face images rather than distant faces",
+            )
+            style = gr.Dropdown(
+                label="Style template",
+                choices=STYLE_NAMES,
+                value=DEFAULT_STYLE_NAME,
+            )
+
+            # strength
+            identitynet_strength_ratio = gr.Slider(
+                label="IdentityNet strength (for fidelity)",
+                minimum=0,
+                maximum=1.5,
+                step=0.05,
+                value=0.80,
+            )
+            adapter_strength_ratio = gr.Slider(
+                label="Image adapter strength (for detail)",
+                minimum=0,
+                maximum=1.5,
+                step=0.05,
+                value=0.80,
+            )
+            with gr.Accordion("Controlnet"):
+                controlnet_selection = gr.CheckboxGroup(
+                    ["pose", "canny", "depth"], label="Controlnet", value=["pose"],
+                    info="Use pose for skeleton inference, canny for edge detection, and depth for depth map estimation. You can try all three to control the generation process"
+                )
+                pose_strength = gr.Slider(
+                    label="Pose strength",
+                    minimum=0,
+                    maximum=1.5,
+                    step=0.05,
+                    value=0.40,
+                )
+                canny_strength = gr.Slider(
+                    label="Canny strength",
+                    minimum=0,
+                    maximum=1.5,
+                    step=0.05,
+                    value=0.40,
+                )
+                depth_strength = gr.Slider(
+                    label="Depth strength",
+                    minimum=0,
+                    maximum=1.5,
+                    step=0.05,
+                    value=0.40,
+                )
+            with gr.Accordion(open=False, label="Advanced Options"):
+                negative_prompt = gr.Textbox(
+                    label="Negative Prompt",
+                    placeholder="low quality",
+                    value="(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
+                )
+                num_steps = gr.Slider(
+                    label="Number of sample steps",
+                    minimum=1,
+                    maximum=100,
+                    step=1,
+                    value=5 if enable_lcm_arg else 30,
+                )
+                guidance_scale = gr.Slider(
+                    label="Guidance scale",
+                    minimum=0.1,
+                    maximum=20.0,
+                    step=0.1,
+                    value=0.0 if enable_lcm_arg else 5.0,
+                )
+                seed = gr.Slider(
+                    label="Seed",
+                    minimum=0,
+                    maximum=MAX_SEED,
+                    step=1,
+                    value=42,
+                )
+                schedulers = [
+                    "DEISMultistepScheduler",
+                    "HeunDiscreteScheduler",
+                    "EulerDiscreteScheduler",
+                    "DPMSolverMultistepScheduler",
+                    "DPMSolverMultistepScheduler-Karras",
+                    "DPMSolverMultistepScheduler-Karras-SDE",
+                ]
+                scheduler = gr.Dropdown(
+                    label="Schedulers",
+                    choices=schedulers,
+                    value="EulerDiscreteScheduler",
+                )
+                randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+                enhance_face_region = gr.Checkbox(label="Enhance non-face region", value=True)
+
+        with gr.Column(scale=1):
+            gallery = gr.Image(label="Generated Images")
+            usage_tips = gr.Markdown(
+                label="InstantID Usage Tips", value=tips, visible=False
+            )
+
+        submit.click(
+            fn=remove_tips,
+            outputs=usage_tips,
+        ).then(
+            fn=randomize_seed_fn,
+            inputs=[seed, randomize_seed],
+            outputs=seed,
+            queue=False,
+            api_name=False,
+        ).then(
+            fn=generate_image,
+            inputs=[
+                face_file,
+                pose_file,
+                prompt,
+                negative_prompt,
+                style,
+                num_steps,
+                identitynet_strength_ratio,
+                adapter_strength_ratio,
+                pose_strength,
+                canny_strength,
+                depth_strength,
+                controlnet_selection,
+                guidance_scale,
+                seed,
+                scheduler,
+                enable_LCM,
+                enhance_face_region,
+            ],
+            outputs=[gallery, usage_tips],
+        )
+
+        enable_LCM.input(
+            fn=toggle_lcm_ui,
+            inputs=[enable_LCM],
+            outputs=[num_steps, guidance_scale],
+            queue=False,
+        )
+
+    gr.Examples(
+        examples=get_example(),
+        inputs=[face_file, pose_file, prompt, style, negative_prompt],
+        fn=run_for_examples,
+        outputs=[gallery, usage_tips],
+        cache_examples=True,
+    )
+
+    gr.Markdown(article)
+
+demo.queue(api_open=False)
+demo.launch()

depth_anything/blocks.py

@@ -0,0 +1,153 @@
+import torch.nn as nn
+
+
+def _make_scratch(in_shape, out_shape, groups=1, expand=False):
+    scratch = nn.Module()
+
+    out_shape1 = out_shape
+    out_shape2 = out_shape
+    out_shape3 = out_shape
+    if len(in_shape) >= 4:
+        out_shape4 = out_shape
+
+    if expand:
+        out_shape1 = out_shape
+        out_shape2 = out_shape*2
+        out_shape3 = out_shape*4
+        if len(in_shape) >= 4:
+            out_shape4 = out_shape*8
+
+    scratch.layer1_rn = nn.Conv2d(
+        in_shape[0], out_shape1, kernel_size=3, stride=1, padding=1, bias=False, groups=groups
+    )
+    scratch.layer2_rn = nn.Conv2d(
+        in_shape[1], out_shape2, kernel_size=3, stride=1, padding=1, bias=False, groups=groups
+    )
+    scratch.layer3_rn = nn.Conv2d(
+        in_shape[2], out_shape3, kernel_size=3, stride=1, padding=1, bias=False, groups=groups
+    )
+    if len(in_shape) >= 4:
+        scratch.layer4_rn = nn.Conv2d(
+            in_shape[3], out_shape4, kernel_size=3, stride=1, padding=1, bias=False, groups=groups
+        )
+
+    return scratch
+
+
+class ResidualConvUnit(nn.Module):
+    """Residual convolution module.
+    """
+
+    def __init__(self, features, activation, bn):
+        """Init.
+
+        Args:
+            features (int): number of features
+        """
+        super().__init__()
+
+        self.bn = bn
+
+        self.groups=1
+
+        self.conv1 = nn.Conv2d(
+            features, features, kernel_size=3, stride=1, padding=1, bias=True, groups=self.groups
+        )
+        
+        self.conv2 = nn.Conv2d(
+            features, features, kernel_size=3, stride=1, padding=1, bias=True, groups=self.groups
+        )
+
+        if self.bn==True:
+            self.bn1 = nn.BatchNorm2d(features)
+            self.bn2 = nn.BatchNorm2d(features)
+
+        self.activation = activation
+
+        self.skip_add = nn.quantized.FloatFunctional()
+
+    def forward(self, x):
+        """Forward pass.
+
+        Args:
+            x (tensor): input
+
+        Returns:
+            tensor: output
+        """
+        
+        out = self.activation(x)
+        out = self.conv1(out)
+        if self.bn==True:
+            out = self.bn1(out)
+       
+        out = self.activation(out)
+        out = self.conv2(out)
+        if self.bn==True:
+            out = self.bn2(out)
+
+        if self.groups > 1:
+            out = self.conv_merge(out)
+
+        return self.skip_add.add(out, x)
+
+
+class FeatureFusionBlock(nn.Module):
+    """Feature fusion block.
+    """
+
+    def __init__(self, features, activation, deconv=False, bn=False, expand=False, align_corners=True, size=None):
+        """Init.
+
+        Args:
+            features (int): number of features
+        """
+        super(FeatureFusionBlock, self).__init__()
+
+        self.deconv = deconv
+        self.align_corners = align_corners
+
+        self.groups=1
+
+        self.expand = expand
+        out_features = features
+        if self.expand==True:
+            out_features = features//2
+        
+        self.out_conv = nn.Conv2d(features, out_features, kernel_size=1, stride=1, padding=0, bias=True, groups=1)
+
+        self.resConfUnit1 = ResidualConvUnit(features, activation, bn)
+        self.resConfUnit2 = ResidualConvUnit(features, activation, bn)
+        
+        self.skip_add = nn.quantized.FloatFunctional()
+
+        self.size=size
+
+    def forward(self, *xs, size=None):
+        """Forward pass.
+
+        Returns:
+            tensor: output
+        """
+        output = xs[0]
+
+        if len(xs) == 2:
+            res = self.resConfUnit1(xs[1])
+            output = self.skip_add.add(output, res)
+
+        output = self.resConfUnit2(output)
+
+        if (size is None) and (self.size is None):
+            modifier = {"scale_factor": 2}
+        elif size is None:
+            modifier = {"size": self.size}
+        else:
+            modifier = {"size": size}
+
+        output = nn.functional.interpolate(
+            output, **modifier, mode="bilinear", align_corners=self.align_corners
+        )
+
+        output = self.out_conv(output)
+
+        return output

depth_anything/dpt.py

@@ -0,0 +1,187 @@
+import argparse
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from huggingface_hub import PyTorchModelHubMixin, hf_hub_download
+
+from depth_anything.blocks import FeatureFusionBlock, _make_scratch
+
+
+def _make_fusion_block(features, use_bn, size = None):
+    return FeatureFusionBlock(
+        features,
+        nn.ReLU(False),
+        deconv=False,
+        bn=use_bn,
+        expand=False,
+        align_corners=True,
+        size=size,
+    )
+
+
+class DPTHead(nn.Module):
+    def __init__(self, nclass, in_channels, features=256, use_bn=False, out_channels=[256, 512, 1024, 1024], use_clstoken=False):
+        super(DPTHead, self).__init__()
+        
+        self.nclass = nclass
+        self.use_clstoken = use_clstoken
+        
+        self.projects = nn.ModuleList([
+            nn.Conv2d(
+                in_channels=in_channels,
+                out_channels=out_channel,
+                kernel_size=1,
+                stride=1,
+                padding=0,
+            ) for out_channel in out_channels
+        ])
+        
+        self.resize_layers = nn.ModuleList([
+            nn.ConvTranspose2d(
+                in_channels=out_channels[0],
+                out_channels=out_channels[0],
+                kernel_size=4,
+                stride=4,
+                padding=0),
+            nn.ConvTranspose2d(
+                in_channels=out_channels[1],
+                out_channels=out_channels[1],
+                kernel_size=2,
+                stride=2,
+                padding=0),
+            nn.Identity(),
+            nn.Conv2d(
+                in_channels=out_channels[3],
+                out_channels=out_channels[3],
+                kernel_size=3,
+                stride=2,
+                padding=1)
+        ])
+        
+        if use_clstoken:
+            self.readout_projects = nn.ModuleList()
+            for _ in range(len(self.projects)):
+                self.readout_projects.append(
+                    nn.Sequential(
+                        nn.Linear(2 * in_channels, in_channels),
+                        nn.GELU()))
+        
+        self.scratch = _make_scratch(
+            out_channels,
+            features,
+            groups=1,
+            expand=False,
+        )
+
+        self.scratch.stem_transpose = None
+        
+        self.scratch.refinenet1 = _make_fusion_block(features, use_bn)
+        self.scratch.refinenet2 = _make_fusion_block(features, use_bn)
+        self.scratch.refinenet3 = _make_fusion_block(features, use_bn)
+        self.scratch.refinenet4 = _make_fusion_block(features, use_bn)
+
+        head_features_1 = features
+        head_features_2 = 32
+        
+        if nclass > 1:
+            self.scratch.output_conv = nn.Sequential(
+                nn.Conv2d(head_features_1, head_features_1, kernel_size=3, stride=1, padding=1),
+                nn.ReLU(True),
+                nn.Conv2d(head_features_1, nclass, kernel_size=1, stride=1, padding=0),
+            )
+        else:
+            self.scratch.output_conv1 = nn.Conv2d(head_features_1, head_features_1 // 2, kernel_size=3, stride=1, padding=1)
+            
+            self.scratch.output_conv2 = nn.Sequential(
+                nn.Conv2d(head_features_1 // 2, head_features_2, kernel_size=3, stride=1, padding=1),
+                nn.ReLU(True),
+                nn.Conv2d(head_features_2, 1, kernel_size=1, stride=1, padding=0),
+                nn.ReLU(True),
+                nn.Identity(),
+            )
+            
+    def forward(self, out_features, patch_h, patch_w):
+        out = []
+        for i, x in enumerate(out_features):
+            if self.use_clstoken:
+                x, cls_token = x[0], x[1]
+                readout = cls_token.unsqueeze(1).expand_as(x)
+                x = self.readout_projects[i](torch.cat((x, readout), -1))
+            else:
+                x = x[0]
+            
+            x = x.permute(0, 2, 1).reshape((x.shape[0], x.shape[-1], patch_h, patch_w))
+            
+            x = self.projects[i](x)
+            x = self.resize_layers[i](x)
+            
+            out.append(x)
+        
+        layer_1, layer_2, layer_3, layer_4 = out
+        
+        layer_1_rn = self.scratch.layer1_rn(layer_1)
+        layer_2_rn = self.scratch.layer2_rn(layer_2)
+        layer_3_rn = self.scratch.layer3_rn(layer_3)
+        layer_4_rn = self.scratch.layer4_rn(layer_4)
+        
+        path_4 = self.scratch.refinenet4(layer_4_rn, size=layer_3_rn.shape[2:])
+        path_3 = self.scratch.refinenet3(path_4, layer_3_rn, size=layer_2_rn.shape[2:])
+        path_2 = self.scratch.refinenet2(path_3, layer_2_rn, size=layer_1_rn.shape[2:])
+        path_1 = self.scratch.refinenet1(path_2, layer_1_rn)
+        
+        out = self.scratch.output_conv1(path_1)
+        out = F.interpolate(out, (int(patch_h * 14), int(patch_w * 14)), mode="bilinear", align_corners=True)
+        out = self.scratch.output_conv2(out)
+        
+        return out
+        
+        
+class DPT_DINOv2(nn.Module):
+    def __init__(self, encoder='vitl', features=256, out_channels=[256, 512, 1024, 1024], use_bn=False, use_clstoken=False, localhub=True):
+        super(DPT_DINOv2, self).__init__()
+        
+        assert encoder in ['vits', 'vitb', 'vitl']
+        
+        # in case the Internet connection is not stable, please load the DINOv2 locally
+        if localhub:
+            self.pretrained = torch.hub.load('torchhub/facebookresearch_dinov2_main', 'dinov2_{:}14'.format(encoder), source='local', pretrained=False)
+        else:
+            self.pretrained = torch.hub.load('facebookresearch/dinov2', 'dinov2_{:}14'.format(encoder))
+        
+        dim = self.pretrained.blocks[0].attn.qkv.in_features
+        
+        self.depth_head = DPTHead(1, dim, features, use_bn, out_channels=out_channels, use_clstoken=use_clstoken)
+        
+    def forward(self, x):
+        h, w = x.shape[-2:]
+        
+        features = self.pretrained.get_intermediate_layers(x, 4, return_class_token=True)
+        
+        patch_h, patch_w = h // 14, w // 14
+
+        depth = self.depth_head(features, patch_h, patch_w)
+        depth = F.interpolate(depth, size=(h, w), mode="bilinear", align_corners=True)
+        depth = F.relu(depth)
+
+        return depth.squeeze(1)
+
+
+class DepthAnything(DPT_DINOv2, PyTorchModelHubMixin):
+    def __init__(self, config):
+        super().__init__(**config)
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--encoder",
+        default="vits",
+        type=str,
+        choices=["vits", "vitb", "vitl"],
+    )
+    args = parser.parse_args()
+    
+    model = DepthAnything.from_pretrained("LiheYoung/depth_anything_{:}14".format(args.encoder))
+    
+    print(model)
+    

depth_anything/util/transform.py

@@ -0,0 +1,248 @@
+import random
+from PIL import Image, ImageOps, ImageFilter
+import torch
+from torchvision import transforms
+import torch.nn.functional as F
+
+import numpy as np
+import cv2
+import math
+
+
+def apply_min_size(sample, size, image_interpolation_method=cv2.INTER_AREA):
+    """Rezise the sample to ensure the given size. Keeps aspect ratio.
+
+    Args:
+        sample (dict): sample
+        size (tuple): image size
+
+    Returns:
+        tuple: new size
+    """
+    shape = list(sample["disparity"].shape)
+
+    if shape[0] >= size[0] and shape[1] >= size[1]:
+        return sample
+
+    scale = [0, 0]
+    scale[0] = size[0] / shape[0]
+    scale[1] = size[1] / shape[1]
+
+    scale = max(scale)
+
+    shape[0] = math.ceil(scale * shape[0])
+    shape[1] = math.ceil(scale * shape[1])
+
+    # resize
+    sample["image"] = cv2.resize(
+        sample["image"], tuple(shape[::-1]), interpolation=image_interpolation_method
+    )
+
+    sample["disparity"] = cv2.resize(
+        sample["disparity"], tuple(shape[::-1]), interpolation=cv2.INTER_NEAREST
+    )
+    sample["mask"] = cv2.resize(
+        sample["mask"].astype(np.float32),
+        tuple(shape[::-1]),
+        interpolation=cv2.INTER_NEAREST,
+    )
+    sample["mask"] = sample["mask"].astype(bool)
+
+    return tuple(shape)
+
+
+class Resize(object):
+    """Resize sample to given size (width, height).
+    """
+
+    def __init__(
+        self,
+        width,
+        height,
+        resize_target=True,
+        keep_aspect_ratio=False,
+        ensure_multiple_of=1,
+        resize_method="lower_bound",
+        image_interpolation_method=cv2.INTER_AREA,
+    ):
+        """Init.
+
+        Args:
+            width (int): desired output width
+            height (int): desired output height
+            resize_target (bool, optional):
+                True: Resize the full sample (image, mask, target).
+                False: Resize image only.
+                Defaults to True.
+            keep_aspect_ratio (bool, optional):
+                True: Keep the aspect ratio of the input sample.
+                Output sample might not have the given width and height, and
+                resize behaviour depends on the parameter 'resize_method'.
+                Defaults to False.
+            ensure_multiple_of (int, optional):
+                Output width and height is constrained to be multiple of this parameter.
+                Defaults to 1.
+            resize_method (str, optional):
+                "lower_bound": Output will be at least as large as the given size.
+                "upper_bound": Output will be at max as large as the given size. (Output size might be smaller than given size.)
+                "minimal": Scale as least as possible.  (Output size might be smaller than given size.)
+                Defaults to "lower_bound".
+        """
+        self.__width = width
+        self.__height = height
+
+        self.__resize_target = resize_target
+        self.__keep_aspect_ratio = keep_aspect_ratio
+        self.__multiple_of = ensure_multiple_of
+        self.__resize_method = resize_method
+        self.__image_interpolation_method = image_interpolation_method
+
+    def constrain_to_multiple_of(self, x, min_val=0, max_val=None):
+        y = (np.round(x / self.__multiple_of) * self.__multiple_of).astype(int)
+
+        if max_val is not None and y > max_val:
+            y = (np.floor(x / self.__multiple_of) * self.__multiple_of).astype(int)
+
+        if y < min_val:
+            y = (np.ceil(x / self.__multiple_of) * self.__multiple_of).astype(int)
+
+        return y
+
+    def get_size(self, width, height):
+        # determine new height and width
+        scale_height = self.__height / height
+        scale_width = self.__width / width
+
+        if self.__keep_aspect_ratio:
+            if self.__resize_method == "lower_bound":
+                # scale such that output size is lower bound
+                if scale_width > scale_height:
+                    # fit width
+                    scale_height = scale_width
+                else:
+                    # fit height
+                    scale_width = scale_height
+            elif self.__resize_method == "upper_bound":
+                # scale such that output size is upper bound
+                if scale_width < scale_height:
+                    # fit width
+                    scale_height = scale_width
+                else:
+                    # fit height
+                    scale_width = scale_height
+            elif self.__resize_method == "minimal":
+                # scale as least as possbile
+                if abs(1 - scale_width) < abs(1 - scale_height):
+                    # fit width
+                    scale_height = scale_width
+                else:
+                    # fit height
+                    scale_width = scale_height
+            else:
+                raise ValueError(
+                    f"resize_method {self.__resize_method} not implemented"
+                )
+
+        if self.__resize_method == "lower_bound":
+            new_height = self.constrain_to_multiple_of(
+                scale_height * height, min_val=self.__height
+            )
+            new_width = self.constrain_to_multiple_of(
+                scale_width * width, min_val=self.__width
+            )
+        elif self.__resize_method == "upper_bound":
+            new_height = self.constrain_to_multiple_of(
+                scale_height * height, max_val=self.__height
+            )
+            new_width = self.constrain_to_multiple_of(
+                scale_width * width, max_val=self.__width
+            )
+        elif self.__resize_method == "minimal":
+            new_height = self.constrain_to_multiple_of(scale_height * height)
+            new_width = self.constrain_to_multiple_of(scale_width * width)
+        else:
+            raise ValueError(f"resize_method {self.__resize_method} not implemented")
+
+        return (new_width, new_height)
+
+    def __call__(self, sample):
+        width, height = self.get_size(
+            sample["image"].shape[1], sample["image"].shape[0]
+        )
+
+        # resize sample
+        sample["image"] = cv2.resize(
+            sample["image"],
+            (width, height),
+            interpolation=self.__image_interpolation_method,
+        )
+
+        if self.__resize_target:
+            if "disparity" in sample:
+                sample["disparity"] = cv2.resize(
+                    sample["disparity"],
+                    (width, height),
+                    interpolation=cv2.INTER_NEAREST,
+                )
+
+            if "depth" in sample:
+                sample["depth"] = cv2.resize(
+                    sample["depth"], (width, height), interpolation=cv2.INTER_NEAREST
+                )
+
+            if "semseg_mask" in sample:
+                # sample["semseg_mask"] = cv2.resize(
+                #     sample["semseg_mask"], (width, height), interpolation=cv2.INTER_NEAREST
+                # )
+                sample["semseg_mask"] = F.interpolate(torch.from_numpy(sample["semseg_mask"]).float()[None, None, ...], (height, width), mode='nearest').numpy()[0, 0]
+                
+            if "mask" in sample:
+                sample["mask"] = cv2.resize(
+                    sample["mask"].astype(np.float32),
+                    (width, height),
+                    interpolation=cv2.INTER_NEAREST,
+                )
+                # sample["mask"] = sample["mask"].astype(bool)
+
+        # print(sample['image'].shape, sample['depth'].shape)
+        return sample
+
+
+class NormalizeImage(object):
+    """Normlize image by given mean and std.
+    """
+
+    def __init__(self, mean, std):
+        self.__mean = mean
+        self.__std = std
+
+    def __call__(self, sample):
+        sample["image"] = (sample["image"] - self.__mean) / self.__std
+
+        return sample
+
+
+class PrepareForNet(object):
+    """Prepare sample for usage as network input.
+    """
+
+    def __init__(self):
+        pass
+
+    def __call__(self, sample):
+        image = np.transpose(sample["image"], (2, 0, 1))
+        sample["image"] = np.ascontiguousarray(image).astype(np.float32)
+
+        if "mask" in sample:
+            sample["mask"] = sample["mask"].astype(np.float32)
+            sample["mask"] = np.ascontiguousarray(sample["mask"])
+        
+        if "depth" in sample:
+            depth = sample["depth"].astype(np.float32)
+            sample["depth"] = np.ascontiguousarray(depth)
+            
+        if "semseg_mask" in sample:
+            sample["semseg_mask"] = sample["semseg_mask"].astype(np.float32)
+            sample["semseg_mask"] = np.ascontiguousarray(sample["semseg_mask"])
+
+        return sample

ip_adapter/attention_processor.py

@@ -0,0 +1,446 @@
+# modified from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+try:
+    import xformers
+    import xformers.ops
+    xformers_available = True
+except Exception as e:
+    xformers_available = False
+
+class RegionControler(object):
+    def __init__(self) -> None:
+        self.prompt_image_conditioning = []
+region_control = RegionControler()
+
+class AttnProcessor(nn.Module):
+    r"""
+    Default processor for performing attention-related computations.
+    """
+    def __init__(
+        self,
+        hidden_size=None,
+        cross_attention_dim=None,
+    ):
+        super().__init__()
+
+    def forward(
+        self,
+        attn,
+        hidden_states,
+        encoder_hidden_states=None,
+        attention_mask=None,
+        temb=None,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        elif attn.norm_cross:
+            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states)
+
+        query = attn.head_to_batch_dim(query)
+        key = attn.head_to_batch_dim(key)
+        value = attn.head_to_batch_dim(value)
+
+        attention_probs = attn.get_attention_scores(query, key, attention_mask)
+        hidden_states = torch.bmm(attention_probs, value)
+        hidden_states = attn.batch_to_head_dim(hidden_states)
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+    
+    
+class IPAttnProcessor(nn.Module):
+    r"""
+    Attention processor for IP-Adapater.
+    Args:
+        hidden_size (`int`):
+            The hidden size of the attention layer.
+        cross_attention_dim (`int`):
+            The number of channels in the `encoder_hidden_states`.
+        scale (`float`, defaults to 1.0):
+            the weight scale of image prompt.
+        num_tokens (`int`, defaults to 4 when do ip_adapter_plus it should be 16):
+            The context length of the image features.
+    """
+
+    def __init__(self, hidden_size, cross_attention_dim=None, scale=1.0, num_tokens=4):
+        super().__init__()
+
+        self.hidden_size = hidden_size
+        self.cross_attention_dim = cross_attention_dim
+        self.scale = scale
+        self.num_tokens = num_tokens
+
+        self.to_k_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
+        self.to_v_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
+
+    def forward(
+        self,
+        attn,
+        hidden_states,
+        encoder_hidden_states=None,
+        attention_mask=None,
+        temb=None,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        else:
+            # get encoder_hidden_states, ip_hidden_states
+            end_pos = encoder_hidden_states.shape[1] - self.num_tokens
+            encoder_hidden_states, ip_hidden_states = encoder_hidden_states[:, :end_pos, :], encoder_hidden_states[:, end_pos:, :]
+            if attn.norm_cross:
+                encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states)
+
+        query = attn.head_to_batch_dim(query)
+        key = attn.head_to_batch_dim(key)
+        value = attn.head_to_batch_dim(value)
+
+        if xformers_available:
+            hidden_states = self._memory_efficient_attention_xformers(query, key, value, attention_mask)
+        else:
+            attention_probs = attn.get_attention_scores(query, key, attention_mask)
+            hidden_states = torch.bmm(attention_probs, value)
+        hidden_states = attn.batch_to_head_dim(hidden_states)
+        
+        # for ip-adapter
+        ip_key = self.to_k_ip(ip_hidden_states)
+        ip_value = self.to_v_ip(ip_hidden_states)
+        
+        ip_key = attn.head_to_batch_dim(ip_key)
+        ip_value = attn.head_to_batch_dim(ip_value)
+        
+        if xformers_available:
+            ip_hidden_states = self._memory_efficient_attention_xformers(query, ip_key, ip_value, None)
+        else:
+            ip_attention_probs = attn.get_attention_scores(query, ip_key, None)
+            ip_hidden_states = torch.bmm(ip_attention_probs, ip_value)
+        ip_hidden_states = attn.batch_to_head_dim(ip_hidden_states)
+
+        # region control
+        if len(region_control.prompt_image_conditioning) == 1:
+            region_mask = region_control.prompt_image_conditioning[0].get('region_mask', None)
+            if region_mask is not None:
+                h, w = region_mask.shape[:2]
+                ratio = (h * w / query.shape[1]) ** 0.5
+                mask = F.interpolate(region_mask[None, None], scale_factor=1/ratio, mode='nearest').reshape([1, -1, 1])
+            else:
+                mask = torch.ones_like(ip_hidden_states)
+            ip_hidden_states = ip_hidden_states * mask     
+
+        hidden_states = hidden_states + self.scale * ip_hidden_states
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+
+    def _memory_efficient_attention_xformers(self, query, key, value, attention_mask):
+        # TODO attention_mask
+        query = query.contiguous()
+        key = key.contiguous()
+        value = value.contiguous()
+        hidden_states = xformers.ops.memory_efficient_attention(query, key, value, attn_bias=attention_mask)
+        # hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
+        return hidden_states
+
+
+class AttnProcessor2_0(torch.nn.Module):
+    r"""
+    Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0).
+    """
+    def __init__(
+        self,
+        hidden_size=None,
+        cross_attention_dim=None,
+    ):
+        super().__init__()
+        if not hasattr(F, "scaled_dot_product_attention"):
+            raise ImportError("AttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
+
+    def forward(
+        self,
+        attn,
+        hidden_states,
+        encoder_hidden_states=None,
+        attention_mask=None,
+        temb=None,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+
+        if attention_mask is not None:
+            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+            # scaled_dot_product_attention expects attention_mask shape to be
+            # (batch, heads, source_length, target_length)
+            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        elif attn.norm_cross:
+            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states)
+
+        inner_dim = key.shape[-1]
+        head_dim = inner_dim // attn.heads
+
+        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+        # TODO: add support for attn.scale when we move to Torch 2.1
+        hidden_states = F.scaled_dot_product_attention(
+            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        hidden_states = hidden_states.to(query.dtype)
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+class IPAttnProcessor2_0(torch.nn.Module):
+    r"""
+    Attention processor for IP-Adapater for PyTorch 2.0.
+    Args:
+        hidden_size (`int`):
+            The hidden size of the attention layer.
+        cross_attention_dim (`int`):
+            The number of channels in the `encoder_hidden_states`.
+        scale (`float`, defaults to 1.0):
+            the weight scale of image prompt.
+        num_tokens (`int`, defaults to 4 when do ip_adapter_plus it should be 16):
+            The context length of the image features.
+    """
+
+    def __init__(self, hidden_size, cross_attention_dim=None, scale=1.0, num_tokens=4):
+        super().__init__()
+
+        if not hasattr(F, "scaled_dot_product_attention"):
+            raise ImportError("AttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
+
+        self.hidden_size = hidden_size
+        self.cross_attention_dim = cross_attention_dim
+        self.scale = scale
+        self.num_tokens = num_tokens
+
+        self.to_k_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
+        self.to_v_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
+
+    def forward(
+        self,
+        attn,
+        hidden_states,
+        encoder_hidden_states=None,
+        attention_mask=None,
+        temb=None,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+
+        if attention_mask is not None:
+            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+            # scaled_dot_product_attention expects attention_mask shape to be
+            # (batch, heads, source_length, target_length)
+            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        else:
+            # get encoder_hidden_states, ip_hidden_states
+            end_pos = encoder_hidden_states.shape[1] - self.num_tokens
+            encoder_hidden_states, ip_hidden_states = (
+                encoder_hidden_states[:, :end_pos, :],
+                encoder_hidden_states[:, end_pos:, :],
+            )
+            if attn.norm_cross:
+                encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states)
+
+        inner_dim = key.shape[-1]
+        head_dim = inner_dim // attn.heads
+
+        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+        # TODO: add support for attn.scale when we move to Torch 2.1
+        hidden_states = F.scaled_dot_product_attention(
+            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        hidden_states = hidden_states.to(query.dtype)
+
+        # for ip-adapter
+        ip_key = self.to_k_ip(ip_hidden_states)
+        ip_value = self.to_v_ip(ip_hidden_states)
+
+        ip_key = ip_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        ip_value = ip_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+        # TODO: add support for attn.scale when we move to Torch 2.1
+        ip_hidden_states = F.scaled_dot_product_attention(
+            query, ip_key, ip_value, attn_mask=None, dropout_p=0.0, is_causal=False
+        )
+        with torch.no_grad():
+            self.attn_map = query @ ip_key.transpose(-2, -1).softmax(dim=-1)
+            #print(self.attn_map.shape)
+
+        ip_hidden_states = ip_hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        ip_hidden_states = ip_hidden_states.to(query.dtype)
+
+        # region control
+        if len(region_control.prompt_image_conditioning) == 1:
+            region_mask = region_control.prompt_image_conditioning[0].get('region_mask', None)
+            if region_mask is not None:
+                h, w = region_mask.shape[:2]
+                ratio = (h * w / query.shape[1]) ** 0.5
+                mask = F.interpolate(region_mask[None, None], scale_factor=1/ratio, mode='nearest').reshape([1, -1, 1])
+            else:
+                mask = torch.ones_like(ip_hidden_states)
+            ip_hidden_states = ip_hidden_states * mask
+
+        hidden_states = hidden_states + self.scale * ip_hidden_states
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states

ip_adapter/resampler.py

@@ -0,0 +1,121 @@
+# modified from https://github.com/mlfoundations/open_flamingo/blob/main/open_flamingo/src/helpers.py
+import math
+
+import torch
+import torch.nn as nn
+
+
+# FFN
+def FeedForward(dim, mult=4):
+    inner_dim = int(dim * mult)
+    return nn.Sequential(
+        nn.LayerNorm(dim),
+        nn.Linear(dim, inner_dim, bias=False),
+        nn.GELU(),
+        nn.Linear(inner_dim, dim, bias=False),
+    )
+    
+    
+def reshape_tensor(x, heads):
+    bs, length, width = x.shape
+    #(bs, length, width) --> (bs, length, n_heads, dim_per_head)
+    x = x.view(bs, length, heads, -1)
+    # (bs, length, n_heads, dim_per_head) --> (bs, n_heads, length, dim_per_head)
+    x = x.transpose(1, 2)
+    # (bs, n_heads, length, dim_per_head) --> (bs*n_heads, length, dim_per_head)
+    x = x.reshape(bs, heads, length, -1)
+    return x
+
+
+class PerceiverAttention(nn.Module):
+    def __init__(self, *, dim, dim_head=64, heads=8):
+        super().__init__()
+        self.scale = dim_head**-0.5
+        self.dim_head = dim_head
+        self.heads = heads
+        inner_dim = dim_head * heads
+
+        self.norm1 = nn.LayerNorm(dim)
+        self.norm2 = nn.LayerNorm(dim)
+
+        self.to_q = nn.Linear(dim, inner_dim, bias=False)
+        self.to_kv = nn.Linear(dim, inner_dim * 2, bias=False)
+        self.to_out = nn.Linear(inner_dim, dim, bias=False)
+
+
+    def forward(self, x, latents):
+        """
+        Args:
+            x (torch.Tensor): image features
+                shape (b, n1, D)
+            latent (torch.Tensor): latent features
+                shape (b, n2, D)
+        """
+        x = self.norm1(x)
+        latents = self.norm2(latents)
+        
+        b, l, _ = latents.shape
+
+        q = self.to_q(latents)
+        kv_input = torch.cat((x, latents), dim=-2)
+        k, v = self.to_kv(kv_input).chunk(2, dim=-1)
+        
+        q = reshape_tensor(q, self.heads)
+        k = reshape_tensor(k, self.heads)
+        v = reshape_tensor(v, self.heads)
+
+        # attention
+        scale = 1 / math.sqrt(math.sqrt(self.dim_head))
+        weight = (q * scale) @ (k * scale).transpose(-2, -1) # More stable with f16 than dividing afterwards
+        weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
+        out = weight @ v
+        
+        out = out.permute(0, 2, 1, 3).reshape(b, l, -1)
+
+        return self.to_out(out)
+
+
+class Resampler(nn.Module):
+    def __init__(
+        self,
+        dim=1024,
+        depth=8,
+        dim_head=64,
+        heads=16,
+        num_queries=8,
+        embedding_dim=768,
+        output_dim=1024,
+        ff_mult=4,
+    ):
+        super().__init__()
+        
+        self.latents = nn.Parameter(torch.randn(1, num_queries, dim) / dim**0.5)
+        
+        self.proj_in = nn.Linear(embedding_dim, dim)
+
+        self.proj_out = nn.Linear(dim, output_dim)
+        self.norm_out = nn.LayerNorm(output_dim)
+        
+        self.layers = nn.ModuleList([])
+        for _ in range(depth):
+            self.layers.append(
+                nn.ModuleList(
+                    [
+                        PerceiverAttention(dim=dim, dim_head=dim_head, heads=heads),
+                        FeedForward(dim=dim, mult=ff_mult),
+                    ]
+                )
+            )
+
+    def forward(self, x):
+        
+        latents = self.latents.repeat(x.size(0), 1, 1)
+        
+        x = self.proj_in(x)
+        
+        for attn, ff in self.layers:
+            latents = attn(x, latents) + latents
+            latents = ff(latents) + latents
+            
+        latents = self.proj_out(latents)
+        return self.norm_out(latents)

ip_adapter/utils.py

@@ -0,0 +1,5 @@
+import torch.nn.functional as F
+
+
+def is_torch2_available():
+    return hasattr(F, "scaled_dot_product_attention")

models/antelopev2/1k3d68.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df5c06b8a0c12e422b2ed8947b8869faa4105387f199c477af038aa01f9a45cc
+size 143607619

models/antelopev2/2d106det.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f001b856447c413801ef5c42091ed0cd516fcd21f2d6b79635b1e733a7109dbf
+size 5030888

models/antelopev2/genderage.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fde69b1c810857b88c64a335084f1c3fe8f01246c9a191b48c7bb756d6652fb
+size 1322532

models/antelopev2/glintr100.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ab1d6435d639628a6f3e5008dd4f929edf4c4124b1a7169e1048f9fef534cdf
+size 260665334

models/antelopev2/scrfd_10g_bnkps.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5838f7fe053675b1c7a08b633df49e7af5495cee0493c7dcf6697200b85b5b91
+size 16923827

pipeline_stable_diffusion_xl_instantid_full.py

@@ -0,0 +1,1204 @@
+# Copyright 2024 The InstantX Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import cv2
+import math
+
+import numpy as np
+import PIL.Image
+import torch
+import torch.nn.functional as F
+
+from diffusers.image_processor import PipelineImageInput
+
+from diffusers.models import ControlNetModel
+
+from diffusers.utils import (
+    deprecate,
+    logging,
+    replace_example_docstring,
+)
+from diffusers.utils.torch_utils import is_compiled_module, is_torch_version
+from diffusers.pipelines.stable_diffusion_xl import StableDiffusionXLPipelineOutput
+
+from diffusers import StableDiffusionXLControlNetPipeline
+from diffusers.pipelines.controlnet.multicontrolnet import MultiControlNetModel
+from diffusers.utils.import_utils import is_xformers_available
+
+from ip_adapter.resampler import Resampler
+from ip_adapter.utils import is_torch2_available
+
+from ip_adapter.attention_processor import IPAttnProcessor, AttnProcessor
+from ip_adapter.attention_processor import region_control
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```py
+        >>> # !pip install opencv-python transformers accelerate insightface
+        >>> import diffusers
+        >>> from diffusers.utils import load_image
+        >>> from diffusers.models import ControlNetModel
+
+        >>> import cv2
+        >>> import torch
+        >>> import numpy as np
+        >>> from PIL import Image
+        
+        >>> from insightface.app import FaceAnalysis
+        >>> from pipeline_stable_diffusion_xl_instantid import StableDiffusionXLInstantIDPipeline, draw_kps
+
+        >>> # download 'antelopev2' under ./models
+        >>> app = FaceAnalysis(name='antelopev2', root='./', providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])
+        >>> app.prepare(ctx_id=0, det_size=(640, 640))
+        
+        >>> # download models under ./checkpoints
+        >>> face_adapter = f'./checkpoints/ip-adapter.bin'
+        >>> controlnet_path = f'./checkpoints/ControlNetModel'
+        
+        >>> # load IdentityNet
+        >>> controlnet = ControlNetModel.from_pretrained(controlnet_path, torch_dtype=torch.float16)
+        
+        >>> pipe = StableDiffusionXLInstantIDPipeline.from_pretrained(
+        ...     "stabilityai/stable-diffusion-xl-base-1.0", controlnet=controlnet, torch_dtype=torch.float16
+        ... )
+        >>> pipe.cuda()
+        
+        >>> # load adapter
+        >>> pipe.load_ip_adapter_instantid(face_adapter)
+
+        >>> prompt = "analog film photo of a man. faded film, desaturated, 35mm photo, grainy, vignette, vintage, Kodachrome, Lomography, stained, highly detailed, found footage, masterpiece, best quality"
+        >>> negative_prompt = "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, painting, drawing, illustration, glitch, deformed, mutated, cross-eyed, ugly, disfigured (lowres, low quality, worst quality:1.2), (text:1.2), watermark, painting, drawing, illustration, glitch,deformed, mutated, cross-eyed, ugly, disfigured"
+
+        >>> # load an image
+        >>> image = load_image("your-example.jpg")
+        
+        >>> face_info = app.get(cv2.cvtColor(np.array(face_image), cv2.COLOR_RGB2BGR))[-1]
+        >>> face_emb = face_info['embedding']
+        >>> face_kps = draw_kps(face_image, face_info['kps'])
+        
+        >>> pipe.set_ip_adapter_scale(0.8)
+
+        >>> # generate image
+        >>> image = pipe(
+        ...     prompt, image_embeds=face_emb, image=face_kps, controlnet_conditioning_scale=0.8
+        ... ).images[0]
+        ```
+"""
+
+from transformers import CLIPTokenizer
+from diffusers.pipelines.stable_diffusion_xl import StableDiffusionXLPipeline
+class LongPromptWeight(object):
+    
+    """
+    Copied from https://github.com/huggingface/diffusers/blob/main/examples/community/lpw_stable_diffusion_xl.py
+    """
+    
+    def __init__(self) -> None:
+        pass
+
+    def parse_prompt_attention(self, text):
+        """
+        Parses a string with attention tokens and returns a list of pairs: text and its associated weight.
+        Accepted tokens are:
+        (abc) - increases attention to abc by a multiplier of 1.1
+        (abc:3.12) - increases attention to abc by a multiplier of 3.12
+        [abc] - decreases attention to abc by a multiplier of 1.1
+        \( - literal character '('
+        \[ - literal character '['
+        \) - literal character ')'
+        \] - literal character ']'
+        \\ - literal character '\'
+        anything else - just text
+
+        >>> parse_prompt_attention('normal text')
+        [['normal text', 1.0]]
+        >>> parse_prompt_attention('an (important) word')
+        [['an ', 1.0], ['important', 1.1], [' word', 1.0]]
+        >>> parse_prompt_attention('(unbalanced')
+        [['unbalanced', 1.1]]
+        >>> parse_prompt_attention('\(literal\]')
+        [['(literal]', 1.0]]
+        >>> parse_prompt_attention('(unnecessary)(parens)')
+        [['unnecessaryparens', 1.1]]
+        >>> parse_prompt_attention('a (((house:1.3)) [on] a (hill:0.5), sun, (((sky))).')
+        [['a ', 1.0],
+        ['house', 1.5730000000000004],
+        [' ', 1.1],
+        ['on', 1.0],
+        [' a ', 1.1],
+        ['hill', 0.55],
+        [', sun, ', 1.1],
+        ['sky', 1.4641000000000006],
+        ['.', 1.1]]
+        """
+        import re
+
+        re_attention = re.compile(
+            r"""
+                \\\(|\\\)|\\\[|\\]|\\\\|\\|\(|\[|:([+-]?[.\d]+)\)|
+                \)|]|[^\\()\[\]:]+|:
+            """,
+            re.X,
+        )
+
+        re_break = re.compile(r"\s*\bBREAK\b\s*", re.S)
+
+        res = []
+        round_brackets = []
+        square_brackets = []
+
+        round_bracket_multiplier = 1.1
+        square_bracket_multiplier = 1 / 1.1
+
+        def multiply_range(start_position, multiplier):
+            for p in range(start_position, len(res)):
+                res[p][1] *= multiplier
+
+        for m in re_attention.finditer(text):
+            text = m.group(0)
+            weight = m.group(1)
+
+            if text.startswith("\\"):
+                res.append([text[1:], 1.0])
+            elif text == "(":
+                round_brackets.append(len(res))
+            elif text == "[":
+                square_brackets.append(len(res))
+            elif weight is not None and len(round_brackets) > 0:
+                multiply_range(round_brackets.pop(), float(weight))
+            elif text == ")" and len(round_brackets) > 0:
+                multiply_range(round_brackets.pop(), round_bracket_multiplier)
+            elif text == "]" and len(square_brackets) > 0:
+                multiply_range(square_brackets.pop(), square_bracket_multiplier)
+            else:
+                parts = re.split(re_break, text)
+                for i, part in enumerate(parts):
+                    if i > 0:
+                        res.append(["BREAK", -1])
+                    res.append([part, 1.0])
+
+        for pos in round_brackets:
+            multiply_range(pos, round_bracket_multiplier)
+
+        for pos in square_brackets:
+            multiply_range(pos, square_bracket_multiplier)
+
+        if len(res) == 0:
+            res = [["", 1.0]]
+
+        # merge runs of identical weights
+        i = 0
+        while i + 1 < len(res):
+            if res[i][1] == res[i + 1][1]:
+                res[i][0] += res[i + 1][0]
+                res.pop(i + 1)
+            else:
+                i += 1
+
+        return res
+
+    def get_prompts_tokens_with_weights(self, clip_tokenizer: CLIPTokenizer, prompt: str):
+        """
+        Get prompt token ids and weights, this function works for both prompt and negative prompt
+
+        Args:
+            pipe (CLIPTokenizer)
+                A CLIPTokenizer
+            prompt (str)
+                A prompt string with weights
+
+        Returns:
+            text_tokens (list)
+                A list contains token ids
+            text_weight (list)
+                A list contains the correspodent weight of token ids
+
+        Example:
+            import torch
+            from transformers import CLIPTokenizer
+
+            clip_tokenizer = CLIPTokenizer.from_pretrained(
+                "stablediffusionapi/deliberate-v2"
+                , subfolder = "tokenizer"
+                , dtype = torch.float16
+            )
+
+            token_id_list, token_weight_list = get_prompts_tokens_with_weights(
+                clip_tokenizer = clip_tokenizer
+                ,prompt = "a (red:1.5) cat"*70
+            )
+        """
+        texts_and_weights = self.parse_prompt_attention(prompt)
+        text_tokens, text_weights = [], []
+        for word, weight in texts_and_weights:
+            # tokenize and discard the starting and the ending token
+            token = clip_tokenizer(word, truncation=False).input_ids[1:-1]  # so that tokenize whatever length prompt
+            # the returned token is a 1d list: [320, 1125, 539, 320]
+
+            # merge the new tokens to the all tokens holder: text_tokens
+            text_tokens = [*text_tokens, *token]
+
+            # each token chunk will come with one weight, like ['red cat', 2.0]
+            # need to expand weight for each token.
+            chunk_weights = [weight] * len(token)
+
+            # append the weight back to the weight holder: text_weights
+            text_weights = [*text_weights, *chunk_weights]
+        return text_tokens, text_weights
+
+    def group_tokens_and_weights(self, token_ids: list, weights: list, pad_last_block=False):
+        """
+        Produce tokens and weights in groups and pad the missing tokens
+
+        Args:
+            token_ids (list)
+                The token ids from tokenizer
+            weights (list)
+                The weights list from function get_prompts_tokens_with_weights
+            pad_last_block (bool)
+                Control if fill the last token list to 75 tokens with eos
+        Returns:
+            new_token_ids (2d list)
+            new_weights (2d list)
+
+        Example:
+            token_groups,weight_groups = group_tokens_and_weights(
+                token_ids = token_id_list
+                , weights = token_weight_list
+            )
+        """
+        bos, eos = 49406, 49407
+
+        # this will be a 2d list
+        new_token_ids = []
+        new_weights = []
+        while len(token_ids) >= 75:
+            # get the first 75 tokens
+            head_75_tokens = [token_ids.pop(0) for _ in range(75)]
+            head_75_weights = [weights.pop(0) for _ in range(75)]
+
+            # extract token ids and weights
+            temp_77_token_ids = [bos] + head_75_tokens + [eos]
+            temp_77_weights = [1.0] + head_75_weights + [1.0]
+
+            # add 77 token and weights chunk to the holder list
+            new_token_ids.append(temp_77_token_ids)
+            new_weights.append(temp_77_weights)
+
+        # padding the left
+        if len(token_ids) >= 0:
+            padding_len = 75 - len(token_ids) if pad_last_block else 0
+
+            temp_77_token_ids = [bos] + token_ids + [eos] * padding_len + [eos]
+            new_token_ids.append(temp_77_token_ids)
+
+            temp_77_weights = [1.0] + weights + [1.0] * padding_len + [1.0]
+            new_weights.append(temp_77_weights)
+
+        return new_token_ids, new_weights
+
+    def get_weighted_text_embeddings_sdxl(
+        self,
+        pipe: StableDiffusionXLPipeline,
+        prompt: str = "",
+        prompt_2: str = None,
+        neg_prompt: str = "",
+        neg_prompt_2: str = None,
+        prompt_embeds=None,
+        negative_prompt_embeds=None,
+        pooled_prompt_embeds=None,
+        negative_pooled_prompt_embeds=None,
+        extra_emb=None,
+        extra_emb_alpha=0.6,
+    ):
+        """
+        This function can process long prompt with weights, no length limitation
+        for Stable Diffusion XL
+
+        Args:
+            pipe (StableDiffusionPipeline)
+            prompt (str)
+            prompt_2 (str)
+            neg_prompt (str)
+            neg_prompt_2 (str)
+        Returns:
+            prompt_embeds (torch.Tensor)
+            neg_prompt_embeds (torch.Tensor)
+        """
+        # 
+        if prompt_embeds is not None and \
+            negative_prompt_embeds is not None and \
+            pooled_prompt_embeds is not None and \
+            negative_pooled_prompt_embeds is not None:
+            return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
+
+        if prompt_2:
+            prompt = f"{prompt} {prompt_2}"
+
+        if neg_prompt_2:
+            neg_prompt = f"{neg_prompt} {neg_prompt_2}"
+
+        eos = pipe.tokenizer.eos_token_id
+
+        # tokenizer 1
+        prompt_tokens, prompt_weights = self.get_prompts_tokens_with_weights(pipe.tokenizer, prompt)
+        neg_prompt_tokens, neg_prompt_weights = self.get_prompts_tokens_with_weights(pipe.tokenizer, neg_prompt)
+
+        # tokenizer 2
+        # prompt_tokens_2, prompt_weights_2 = self.get_prompts_tokens_with_weights(pipe.tokenizer_2, prompt)
+        # neg_prompt_tokens_2, neg_prompt_weights_2 = self.get_prompts_tokens_with_weights(pipe.tokenizer_2, neg_prompt)
+        # tokenizer 2 遇到 !! !!!! 等多感叹号和tokenizer 1的效果不一致
+        prompt_tokens_2, prompt_weights_2 = self.get_prompts_tokens_with_weights(pipe.tokenizer, prompt)
+        neg_prompt_tokens_2, neg_prompt_weights_2 = self.get_prompts_tokens_with_weights(pipe.tokenizer, neg_prompt)
+
+        # padding the shorter one for prompt set 1
+        prompt_token_len = len(prompt_tokens)
+        neg_prompt_token_len = len(neg_prompt_tokens)
+
+        if prompt_token_len > neg_prompt_token_len:
+            # padding the neg_prompt with eos token
+            neg_prompt_tokens = neg_prompt_tokens + [eos] * abs(prompt_token_len - neg_prompt_token_len)
+            neg_prompt_weights = neg_prompt_weights + [1.0] * abs(prompt_token_len - neg_prompt_token_len)
+        else:
+            # padding the prompt
+            prompt_tokens = prompt_tokens + [eos] * abs(prompt_token_len - neg_prompt_token_len)
+            prompt_weights = prompt_weights + [1.0] * abs(prompt_token_len - neg_prompt_token_len)
+
+        # padding the shorter one for token set 2
+        prompt_token_len_2 = len(prompt_tokens_2)
+        neg_prompt_token_len_2 = len(neg_prompt_tokens_2)
+
+        if prompt_token_len_2 > neg_prompt_token_len_2:
+            # padding the neg_prompt with eos token
+            neg_prompt_tokens_2 = neg_prompt_tokens_2 + [eos] * abs(prompt_token_len_2 - neg_prompt_token_len_2)
+            neg_prompt_weights_2 = neg_prompt_weights_2 + [1.0] * abs(prompt_token_len_2 - neg_prompt_token_len_2)
+        else:
+            # padding the prompt
+            prompt_tokens_2 = prompt_tokens_2 + [eos] * abs(prompt_token_len_2 - neg_prompt_token_len_2)
+            prompt_weights_2 = prompt_weights + [1.0] * abs(prompt_token_len_2 - neg_prompt_token_len_2)
+
+        embeds = []
+        neg_embeds = []
+
+        prompt_token_groups, prompt_weight_groups = self.group_tokens_and_weights(prompt_tokens.copy(), prompt_weights.copy())
+
+        neg_prompt_token_groups, neg_prompt_weight_groups = self.group_tokens_and_weights(
+            neg_prompt_tokens.copy(), neg_prompt_weights.copy()
+        )
+
+        prompt_token_groups_2, prompt_weight_groups_2 = self.group_tokens_and_weights(
+            prompt_tokens_2.copy(), prompt_weights_2.copy()
+        )
+
+        neg_prompt_token_groups_2, neg_prompt_weight_groups_2 = self.group_tokens_and_weights(
+            neg_prompt_tokens_2.copy(), neg_prompt_weights_2.copy()
+        )
+
+        # get prompt embeddings one by one is not working.
+        for i in range(len(prompt_token_groups)):
+            # get positive prompt embeddings with weights
+            token_tensor = torch.tensor([prompt_token_groups[i]], dtype=torch.long, device=pipe.device)
+            weight_tensor = torch.tensor(prompt_weight_groups[i], dtype=torch.float16, device=pipe.device)
+
+            token_tensor_2 = torch.tensor([prompt_token_groups_2[i]], dtype=torch.long, device=pipe.device)
+
+            # use first text encoder
+            prompt_embeds_1 = pipe.text_encoder(token_tensor.to(pipe.device), output_hidden_states=True)
+            prompt_embeds_1_hidden_states = prompt_embeds_1.hidden_states[-2]
+
+            # use second text encoder
+            prompt_embeds_2 = pipe.text_encoder_2(token_tensor_2.to(pipe.device), output_hidden_states=True)
+            prompt_embeds_2_hidden_states = prompt_embeds_2.hidden_states[-2]
+            pooled_prompt_embeds = prompt_embeds_2[0]
+
+            prompt_embeds_list = [prompt_embeds_1_hidden_states, prompt_embeds_2_hidden_states]
+            token_embedding = torch.concat(prompt_embeds_list, dim=-1).squeeze(0)
+
+            for j in range(len(weight_tensor)):
+                if weight_tensor[j] != 1.0:
+                    token_embedding[j] = (
+                        token_embedding[-1] + (token_embedding[j] - token_embedding[-1]) * weight_tensor[j]
+                    )
+
+            token_embedding = token_embedding.unsqueeze(0)
+            embeds.append(token_embedding)
+
+            # get negative prompt embeddings with weights
+            neg_token_tensor = torch.tensor([neg_prompt_token_groups[i]], dtype=torch.long, device=pipe.device)
+            neg_token_tensor_2 = torch.tensor([neg_prompt_token_groups_2[i]], dtype=torch.long, device=pipe.device)
+            neg_weight_tensor = torch.tensor(neg_prompt_weight_groups[i], dtype=torch.float16, device=pipe.device)
+
+            # use first text encoder
+            neg_prompt_embeds_1 = pipe.text_encoder(neg_token_tensor.to(pipe.device), output_hidden_states=True)
+            neg_prompt_embeds_1_hidden_states = neg_prompt_embeds_1.hidden_states[-2]
+
+            # use second text encoder
+            neg_prompt_embeds_2 = pipe.text_encoder_2(neg_token_tensor_2.to(pipe.device), output_hidden_states=True)
+            neg_prompt_embeds_2_hidden_states = neg_prompt_embeds_2.hidden_states[-2]
+            negative_pooled_prompt_embeds = neg_prompt_embeds_2[0]
+
+            neg_prompt_embeds_list = [neg_prompt_embeds_1_hidden_states, neg_prompt_embeds_2_hidden_states]
+            neg_token_embedding = torch.concat(neg_prompt_embeds_list, dim=-1).squeeze(0)
+
+            for z in range(len(neg_weight_tensor)):
+                if neg_weight_tensor[z] != 1.0:
+                    neg_token_embedding[z] = (
+                        neg_token_embedding[-1] + (neg_token_embedding[z] - neg_token_embedding[-1]) * neg_weight_tensor[z]
+                    )
+
+            neg_token_embedding = neg_token_embedding.unsqueeze(0)
+            neg_embeds.append(neg_token_embedding)
+
+        prompt_embeds = torch.cat(embeds, dim=1)
+        negative_prompt_embeds = torch.cat(neg_embeds, dim=1)
+
+        if extra_emb is not None:
+            extra_emb = extra_emb.to(prompt_embeds.device, dtype=prompt_embeds.dtype) * extra_emb_alpha
+            prompt_embeds = torch.cat([prompt_embeds, extra_emb], 1)
+            negative_prompt_embeds = torch.cat([negative_prompt_embeds, torch.zeros_like(extra_emb)], 1)
+            print(f'fix prompt_embeds, extra_emb_alpha={extra_emb_alpha}')
+
+        return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
+
+    def get_prompt_embeds(self, *args, **kwargs):
+        prompt_embeds, negative_prompt_embeds, _, _ = self.get_weighted_text_embeddings_sdxl(*args, **kwargs)
+        prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+        return prompt_embeds
+
+def draw_kps(image_pil, kps, color_list=[(255,0,0), (0,255,0), (0,0,255), (255,255,0), (255,0,255)]):
+    
+    stickwidth = 4
+    limbSeq = np.array([[0, 2], [1, 2], [3, 2], [4, 2]])
+    kps = np.array(kps)
+
+    w, h = image_pil.size
+    out_img = np.zeros([h, w, 3])
+
+    for i in range(len(limbSeq)):
+        index = limbSeq[i]
+        color = color_list[index[0]]
+
+        x = kps[index][:, 0]
+        y = kps[index][:, 1]
+        length = ((x[0] - x[1]) ** 2 + (y[0] - y[1]) ** 2) ** 0.5
+        angle = math.degrees(math.atan2(y[0] - y[1], x[0] - x[1]))
+        polygon = cv2.ellipse2Poly((int(np.mean(x)), int(np.mean(y))), (int(length / 2), stickwidth), int(angle), 0, 360, 1)
+        out_img = cv2.fillConvexPoly(out_img.copy(), polygon, color)
+    out_img = (out_img * 0.6).astype(np.uint8)
+
+    for idx_kp, kp in enumerate(kps):
+        color = color_list[idx_kp]
+        x, y = kp
+        out_img = cv2.circle(out_img.copy(), (int(x), int(y)), 10, color, -1)
+
+    out_img_pil = PIL.Image.fromarray(out_img.astype(np.uint8))
+    return out_img_pil
+    
+class StableDiffusionXLInstantIDPipeline(StableDiffusionXLControlNetPipeline):
+    
+    def cuda(self, dtype=torch.float16, use_xformers=False):
+        self.to('cuda', dtype)
+        
+        if hasattr(self, 'image_proj_model'):
+            self.image_proj_model.to(self.unet.device).to(self.unet.dtype)
+        
+        if use_xformers:
+            if is_xformers_available():
+                import xformers
+                from packaging import version
+
+                xformers_version = version.parse(xformers.__version__)
+                if xformers_version == version.parse("0.0.16"):
+                    logger.warn(
+                        "xFormers 0.0.16 cannot be used for training in some GPUs. If you observe problems during training, please update xFormers to at least 0.0.17. See https://huggingface.co/docs/diffusers/main/en/optimization/xformers for more details."
+                    )
+                self.enable_xformers_memory_efficient_attention()
+            else:
+                raise ValueError("xformers is not available. Make sure it is installed correctly")
+    
+    def load_ip_adapter_instantid(self, model_ckpt, image_emb_dim=512, num_tokens=16, scale=0.5):     
+        self.set_image_proj_model(model_ckpt, image_emb_dim, num_tokens)
+        self.set_ip_adapter(model_ckpt, num_tokens, scale)
+        
+    def set_image_proj_model(self, model_ckpt, image_emb_dim=512, num_tokens=16):
+        
+        image_proj_model = Resampler(
+            dim=1280,
+            depth=4,
+            dim_head=64,
+            heads=20,
+            num_queries=num_tokens,
+            embedding_dim=image_emb_dim,
+            output_dim=self.unet.config.cross_attention_dim,
+            ff_mult=4,
+        )
+
+        image_proj_model.eval()
+        
+        self.image_proj_model = image_proj_model.to(self.device, dtype=self.dtype)
+        state_dict = torch.load(model_ckpt, map_location="cpu")
+        if 'image_proj' in state_dict:
+            state_dict = state_dict["image_proj"]
+        self.image_proj_model.load_state_dict(state_dict)
+        
+        self.image_proj_model_in_features = image_emb_dim
+    
+    def set_ip_adapter(self, model_ckpt, num_tokens, scale):
+        
+        unet = self.unet
+        attn_procs = {}
+        for name in unet.attn_processors.keys():
+            cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
+            if name.startswith("mid_block"):
+                hidden_size = unet.config.block_out_channels[-1]
+            elif name.startswith("up_blocks"):
+                block_id = int(name[len("up_blocks.")])
+                hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
+            elif name.startswith("down_blocks"):
+                block_id = int(name[len("down_blocks.")])
+                hidden_size = unet.config.block_out_channels[block_id]
+            if cross_attention_dim is None:
+                attn_procs[name] = AttnProcessor().to(unet.device, dtype=unet.dtype)
+            else:
+                attn_procs[name] = IPAttnProcessor(hidden_size=hidden_size, 
+                                                   cross_attention_dim=cross_attention_dim, 
+                                                   scale=scale,
+                                                   num_tokens=num_tokens).to(unet.device, dtype=unet.dtype)
+        unet.set_attn_processor(attn_procs)
+        
+        state_dict = torch.load(model_ckpt, map_location="cpu")
+        ip_layers = torch.nn.ModuleList(self.unet.attn_processors.values())
+        if 'ip_adapter' in state_dict:
+            state_dict = state_dict['ip_adapter']
+        ip_layers.load_state_dict(state_dict)
+    
+    def set_ip_adapter_scale(self, scale):
+        unet = getattr(self, self.unet_name) if not hasattr(self, "unet") else self.unet
+        for attn_processor in unet.attn_processors.values():
+            if isinstance(attn_processor, IPAttnProcessor):
+                attn_processor.scale = scale
+
+    def _encode_prompt_image_emb(self, prompt_image_emb, device, num_images_per_prompt, dtype, do_classifier_free_guidance):
+        
+        if isinstance(prompt_image_emb, torch.Tensor):
+            prompt_image_emb = prompt_image_emb.clone().detach()
+        else:
+            prompt_image_emb = torch.tensor(prompt_image_emb)
+            
+        prompt_image_emb = prompt_image_emb.to(device=device, dtype=dtype)
+        prompt_image_emb = prompt_image_emb.reshape([1, -1, self.image_proj_model_in_features])
+        
+        if do_classifier_free_guidance:
+            prompt_image_emb = torch.cat([torch.zeros_like(prompt_image_emb), prompt_image_emb], dim=0)
+        else:
+            prompt_image_emb = torch.cat([prompt_image_emb], dim=0)
+        
+        prompt_image_emb = self.image_proj_model(prompt_image_emb)
+
+        bs_embed, seq_len, _ = prompt_image_emb.shape
+        prompt_image_emb = prompt_image_emb.repeat(1, num_images_per_prompt, 1)
+        prompt_image_emb = prompt_image_emb.view(bs_embed * num_images_per_prompt, seq_len, -1)
+        
+        return prompt_image_emb
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
+        prompt_2: Optional[Union[str, List[str]]] = None,
+        image: PipelineImageInput = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_inference_steps: int = 50,
+        guidance_scale: float = 5.0,
+        negative_prompt: Optional[Union[str, List[str]]] = None,
+        negative_prompt_2: Optional[Union[str, List[str]]] = None,
+        num_images_per_prompt: Optional[int] = 1,
+        eta: float = 0.0,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
+        image_embeds: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+        controlnet_conditioning_scale: Union[float, List[float]] = 1.0,
+        guess_mode: bool = False,
+        control_guidance_start: Union[float, List[float]] = 0.0,
+        control_guidance_end: Union[float, List[float]] = 1.0,
+        original_size: Tuple[int, int] = None,
+        crops_coords_top_left: Tuple[int, int] = (0, 0),
+        target_size: Tuple[int, int] = None,
+        negative_original_size: Optional[Tuple[int, int]] = None,
+        negative_crops_coords_top_left: Tuple[int, int] = (0, 0),
+        negative_target_size: Optional[Tuple[int, int]] = None,
+        clip_skip: Optional[int] = None,
+        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+
+        # IP adapter
+        ip_adapter_scale=None,
+
+        # Enhance Face Region
+        control_mask = None,
+
+        **kwargs,
+    ):
+        r"""
+        The call function to the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+            prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+                used in both text-encoders.
+            image (`torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.FloatTensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+                    `List[List[torch.FloatTensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+                The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
+                specified as `torch.FloatTensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be
+                accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height
+                and/or width are passed, `image` is resized accordingly. If multiple ControlNets are specified in
+                `init`, images must be passed as a list such that each element of the list can be correctly batched for
+                input to a single ControlNet.
+            height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+                The height in pixels of the generated image. Anything below 512 pixels won't work well for
+                [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+                and checkpoints that are not specifically fine-tuned on low resolutions.
+            width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+                The width in pixels of the generated image. Anything below 512 pixels won't work well for
+                [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+                and checkpoints that are not specifically fine-tuned on low resolutions.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            guidance_scale (`float`, *optional*, defaults to 5.0):
+                A higher guidance scale value encourages the model to generate images closely linked to the text
+                `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide what to not include in image generation. If not defined, you need to
+                pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
+            negative_prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide what to not include in image generation. This is sent to `tokenizer_2`
+                and `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders.
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
+                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+                generation deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor is generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+                provided, text embeddings are generated from the `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+                not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+                not provided, pooled text embeddings are generated from `prompt` input argument.
+            negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs (prompt
+                weighting). If not provided, pooled `negative_prompt_embeds` are generated from `negative_prompt` input
+                argument.
+            image_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated image embeddings.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            cross_attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
+                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0):
+                The outputs of the ControlNet are multiplied by `controlnet_conditioning_scale` before they are added
+                to the residual in the original `unet`. If multiple ControlNets are specified in `init`, you can set
+                the corresponding scale as a list.
+            guess_mode (`bool`, *optional*, defaults to `False`):
+                The ControlNet encoder tries to recognize the content of the input image even if you remove all
+                prompts. A `guidance_scale` value between 3.0 and 5.0 is recommended.
+            control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0):
+                The percentage of total steps at which the ControlNet starts applying.
+            control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0):
+                The percentage of total steps at which the ControlNet stops applying.
+            original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+                If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
+                `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as
+                explained in section 2.2 of
+                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+            crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+                `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
+                `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
+                `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
+                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+            target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+                For most cases, `target_size` should be set to the desired height and width of the generated image. If
+                not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in
+                section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+            negative_original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+                To negatively condition the generation process based on a specific image resolution. Part of SDXL's
+                micro-conditioning as explained in section 2.2 of
+                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+                information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+            negative_crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+                To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's
+                micro-conditioning as explained in section 2.2 of
+                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+                information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+            negative_target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+                To negatively condition the generation process based on a target image resolution. It should be as same
+                as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of
+                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+                information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+            clip_skip (`int`, *optional*):
+                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+                the output of the pre-final layer will be used for computing the prompt embeddings.
+            callback_on_step_end (`Callable`, *optional*):
+                A function that calls at the end of each denoising steps during the inference. The function is called
+                with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
+                callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
+                `callback_on_step_end_tensor_inputs`.
+            callback_on_step_end_tensor_inputs (`List`, *optional*):
+                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+                `._callback_tensor_inputs` attribute of your pipeine class.
+
+        Examples:
+
+        Returns:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned,
+                otherwise a `tuple` is returned containing the output images.
+        """
+
+        lpw = LongPromptWeight()
+
+        callback = kwargs.pop("callback", None)
+        callback_steps = kwargs.pop("callback_steps", None)
+
+        if callback is not None:
+            deprecate(
+                "callback",
+                "1.0.0",
+                "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+            )
+        if callback_steps is not None:
+            deprecate(
+                "callback_steps",
+                "1.0.0",
+                "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+            )
+
+        controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+        # align format for control guidance
+        if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list):
+            control_guidance_start = len(control_guidance_end) * [control_guidance_start]
+        elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list):
+            control_guidance_end = len(control_guidance_start) * [control_guidance_end]
+        elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list):
+            mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetModel) else 1
+            control_guidance_start, control_guidance_end = (
+                mult * [control_guidance_start],
+                mult * [control_guidance_end],
+            )
+        
+        # 0. set ip_adapter_scale
+        if ip_adapter_scale is not None:
+            self.set_ip_adapter_scale(ip_adapter_scale)
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            prompt_2,
+            image,
+            callback_steps,
+            negative_prompt,
+            negative_prompt_2,
+            prompt_embeds,
+            negative_prompt_embeds,
+            pooled_prompt_embeds,
+            negative_pooled_prompt_embeds,
+            controlnet_conditioning_scale,
+            control_guidance_start,
+            control_guidance_end,
+            callback_on_step_end_tensor_inputs,
+        )
+
+        self._guidance_scale = guidance_scale
+        self._clip_skip = clip_skip
+        self._cross_attention_kwargs = cross_attention_kwargs
+
+        # 2. Define call parameters
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        device = self._execution_device
+
+        if isinstance(controlnet, MultiControlNetModel) and isinstance(controlnet_conditioning_scale, float):
+            controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets)
+
+        global_pool_conditions = (
+            controlnet.config.global_pool_conditions
+            if isinstance(controlnet, ControlNetModel)
+            else controlnet.nets[0].config.global_pool_conditions
+        )
+        guess_mode = guess_mode or global_pool_conditions
+
+        # 3.1 Encode input prompt
+        (
+            prompt_embeds,
+            negative_prompt_embeds,
+            pooled_prompt_embeds,
+            negative_pooled_prompt_embeds,
+        ) = lpw.get_weighted_text_embeddings_sdxl(
+            pipe=self, 
+            prompt=prompt, 
+            neg_prompt=negative_prompt,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+        )
+        
+        # 3.2 Encode image prompt
+        prompt_image_emb = self._encode_prompt_image_emb(image_embeds, 
+                                                         device,
+                                                         num_images_per_prompt,
+                                                         self.unet.dtype,
+                                                         self.do_classifier_free_guidance)
+        
+        # 4. Prepare image
+        if isinstance(controlnet, ControlNetModel):
+            image = self.prepare_image(
+                image=image,
+                width=width,
+                height=height,
+                batch_size=batch_size * num_images_per_prompt,
+                num_images_per_prompt=num_images_per_prompt,
+                device=device,
+                dtype=controlnet.dtype,
+                do_classifier_free_guidance=self.do_classifier_free_guidance,
+                guess_mode=guess_mode,
+            )
+            height, width = image.shape[-2:]
+        elif isinstance(controlnet, MultiControlNetModel):
+            images = []
+
+            for image_ in image:
+                image_ = self.prepare_image(
+                    image=image_,
+                    width=width,
+                    height=height,
+                    batch_size=batch_size * num_images_per_prompt,
+                    num_images_per_prompt=num_images_per_prompt,
+                    device=device,
+                    dtype=controlnet.dtype,
+                    do_classifier_free_guidance=self.do_classifier_free_guidance,
+                    guess_mode=guess_mode,
+                )
+
+                images.append(image_)
+
+            image = images
+            height, width = image[0].shape[-2:]
+        else:
+            assert False
+
+        # 4.1 Region control
+        if control_mask is not None:
+            mask_weight_image = control_mask
+            mask_weight_image = np.array(mask_weight_image)
+            mask_weight_image_tensor = torch.from_numpy(mask_weight_image).to(device=device, dtype=prompt_embeds.dtype)
+            mask_weight_image_tensor = mask_weight_image_tensor[:, :, 0] / 255.
+            mask_weight_image_tensor = mask_weight_image_tensor[None, None]
+            h, w = mask_weight_image_tensor.shape[-2:]
+            control_mask_wight_image_list = []
+            for scale in [8, 8, 8, 16, 16, 16, 32, 32, 32]:
+                scale_mask_weight_image_tensor = F.interpolate(
+                    mask_weight_image_tensor,(h // scale, w // scale), mode='bilinear')
+                control_mask_wight_image_list.append(scale_mask_weight_image_tensor)
+            region_mask = torch.from_numpy(np.array(control_mask)[:, :, 0]).to(self.unet.device, dtype=self.unet.dtype) / 255.
+            region_control.prompt_image_conditioning = [dict(region_mask=region_mask)]
+        else:
+            control_mask_wight_image_list = None
+            region_control.prompt_image_conditioning = [dict(region_mask=None)]
+
+        # 5. Prepare timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=device)
+        timesteps = self.scheduler.timesteps
+        self._num_timesteps = len(timesteps)
+
+        # 6. Prepare latent variables
+        num_channels_latents = self.unet.config.in_channels
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            generator,
+            latents,
+        )
+
+        # 6.5 Optionally get Guidance Scale Embedding
+        timestep_cond = None
+        if self.unet.config.time_cond_proj_dim is not None:
+            guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt)
+            timestep_cond = self.get_guidance_scale_embedding(
+                guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim
+            ).to(device=device, dtype=latents.dtype)
+
+        # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+        # 7.1 Create tensor stating which controlnets to keep
+        controlnet_keep = []
+        for i in range(len(timesteps)):
+            keeps = [
+                1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e)
+                for s, e in zip(control_guidance_start, control_guidance_end)
+            ]
+            controlnet_keep.append(keeps[0] if isinstance(controlnet, ControlNetModel) else keeps)
+
+        # 7.2 Prepare added time ids & embeddings
+        if isinstance(image, list):
+            original_size = original_size or image[0].shape[-2:]
+        else:
+            original_size = original_size or image.shape[-2:]
+        target_size = target_size or (height, width)
+
+        add_text_embeds = pooled_prompt_embeds
+        if self.text_encoder_2 is None:
+            text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1])
+        else:
+            text_encoder_projection_dim = self.text_encoder_2.config.projection_dim
+
+        add_time_ids = self._get_add_time_ids(
+            original_size,
+            crops_coords_top_left,
+            target_size,
+            dtype=prompt_embeds.dtype,
+            text_encoder_projection_dim=text_encoder_projection_dim,
+        )
+
+        if negative_original_size is not None and negative_target_size is not None:
+            negative_add_time_ids = self._get_add_time_ids(
+                negative_original_size,
+                negative_crops_coords_top_left,
+                negative_target_size,
+                dtype=prompt_embeds.dtype,
+                text_encoder_projection_dim=text_encoder_projection_dim,
+            )
+        else:
+            negative_add_time_ids = add_time_ids
+
+        if self.do_classifier_free_guidance:
+            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+            add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0)
+            add_time_ids = torch.cat([negative_add_time_ids, add_time_ids], dim=0)
+
+        prompt_embeds = prompt_embeds.to(device)
+        add_text_embeds = add_text_embeds.to(device)
+        add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1)
+        encoder_hidden_states = torch.cat([prompt_embeds, prompt_image_emb], dim=1)
+
+        # 8. Denoising loop
+        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+        is_unet_compiled = is_compiled_module(self.unet)
+        is_controlnet_compiled = is_compiled_module(self.controlnet)
+        is_torch_higher_equal_2_1 = is_torch_version(">=", "2.1")
+                
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                # Relevant thread:
+                # https://dev-discuss.pytorch.org/t/cudagraphs-in-pytorch-2-0/1428
+                if (is_unet_compiled and is_controlnet_compiled) and is_torch_higher_equal_2_1:
+                    torch._inductor.cudagraph_mark_step_begin()
+                # expand the latents if we are doing classifier free guidance
+                latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
+                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+                added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids}
+
+                # controlnet(s) inference
+                if guess_mode and self.do_classifier_free_guidance:
+                    # Infer ControlNet only for the conditional batch.
+                    control_model_input = latents
+                    control_model_input = self.scheduler.scale_model_input(control_model_input, t)
+                    controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
+                    controlnet_added_cond_kwargs = {
+                        "text_embeds": add_text_embeds.chunk(2)[1],
+                        "time_ids": add_time_ids.chunk(2)[1],
+                    }
+                else:
+                    control_model_input = latent_model_input
+                    controlnet_prompt_embeds = prompt_embeds
+                    controlnet_added_cond_kwargs = added_cond_kwargs
+                
+                if isinstance(controlnet_keep[i], list):
+                    cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])]
+                else:
+                    controlnet_cond_scale = controlnet_conditioning_scale
+                    if isinstance(controlnet_cond_scale, list):
+                        controlnet_cond_scale = controlnet_cond_scale[0]
+                    cond_scale = controlnet_cond_scale * controlnet_keep[i]
+
+                if isinstance(self.controlnet, MultiControlNetModel):
+                    down_block_res_samples_list, mid_block_res_sample_list = [], []
+                    for control_index in range(len(self.controlnet.nets)):
+                        controlnet = self.controlnet.nets[control_index]
+                        if control_index == 0:
+                            # assume fhe first controlnet is IdentityNet
+                            controlnet_prompt_embeds = prompt_image_emb
+                        else:
+                            controlnet_prompt_embeds = prompt_embeds
+                        down_block_res_samples, mid_block_res_sample = controlnet(control_model_input,
+                                                                                  t,
+                                                                                  encoder_hidden_states=controlnet_prompt_embeds,
+                                                                                  controlnet_cond=image[control_index],
+                                                                                  conditioning_scale=cond_scale[control_index],
+                                                                                  guess_mode=guess_mode,
+                                                                                  added_cond_kwargs=controlnet_added_cond_kwargs,
+                                                                                  return_dict=False)
+
+                        # controlnet mask
+                        if control_index == 0 and control_mask_wight_image_list is not None:
+                            down_block_res_samples = [
+                                down_block_res_sample * mask_weight
+                                for down_block_res_sample, mask_weight in zip(down_block_res_samples, control_mask_wight_image_list)
+                            ]
+                            mid_block_res_sample *= control_mask_wight_image_list[-1]
+
+                        down_block_res_samples_list.append(down_block_res_samples)
+                        mid_block_res_sample_list.append(mid_block_res_sample)
+
+                    mid_block_res_sample = torch.stack(mid_block_res_sample_list).sum(dim=0)
+                    down_block_res_samples = [torch.stack(down_block_res_samples).sum(dim=0) for down_block_res_samples in
+                                              zip(*down_block_res_samples_list)]
+                else:
+                    down_block_res_samples, mid_block_res_sample = self.controlnet(
+                        control_model_input,
+                        t,
+                        encoder_hidden_states=prompt_image_emb,
+                        controlnet_cond=image,
+                        conditioning_scale=cond_scale,
+                        guess_mode=guess_mode,
+                        added_cond_kwargs=controlnet_added_cond_kwargs,
+                        return_dict=False,
+                    )
+
+                    # controlnet mask
+                    if control_mask_wight_image_list is not None:
+                        down_block_res_samples = [
+                            down_block_res_sample * mask_weight
+                            for down_block_res_sample, mask_weight in zip(down_block_res_samples, control_mask_wight_image_list)
+                        ]
+                        mid_block_res_sample *= control_mask_wight_image_list[-1]
+
+                if guess_mode and self.do_classifier_free_guidance:
+                    # Infered ControlNet only for the conditional batch.
+                    # To apply the output of ControlNet to both the unconditional and conditional batches,
+                    # add 0 to the unconditional batch to keep it unchanged.
+                    down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples]
+                    mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])
+
+                # predict the noise residual
+                noise_pred = self.unet(
+                    latent_model_input,
+                    t,
+                    encoder_hidden_states=encoder_hidden_states,
+                    timestep_cond=timestep_cond,
+                    cross_attention_kwargs=self.cross_attention_kwargs,
+                    down_block_additional_residuals=down_block_res_samples,
+                    mid_block_additional_residual=mid_block_res_sample,
+                    added_cond_kwargs=added_cond_kwargs,
+                    return_dict=False,
+                )[0]
+
+                # perform guidance
+                if self.do_classifier_free_guidance:
+                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+                # compute the previous noisy sample x_t -> x_t-1
+                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+                if callback_on_step_end is not None:
+                    callback_kwargs = {}
+                    for k in callback_on_step_end_tensor_inputs:
+                        callback_kwargs[k] = locals()[k]
+                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+                    latents = callback_outputs.pop("latents", latents)
+                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+                    negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+                    if callback is not None and i % callback_steps == 0:
+                        step_idx = i // getattr(self.scheduler, "order", 1)
+                        callback(step_idx, t, latents)
+        
+        if not output_type == "latent":
+            # make sure the VAE is in float32 mode, as it overflows in float16
+            needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast
+            if needs_upcasting:
+                self.upcast_vae()
+                latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+            
+            image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
+
+            # cast back to fp16 if needed
+            if needs_upcasting:
+                self.vae.to(dtype=torch.float16)            
+        else:
+            image = latents
+
+        if not output_type == "latent":
+            # apply watermark if available
+            if self.watermark is not None:
+                image = self.watermark.apply_watermark(image)
+
+            image = self.image_processor.postprocess(image, output_type=output_type)
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (image,)
+
+        return StableDiffusionXLPipelineOutput(images=image)

requirements.txt

@@ -0,0 +1,19 @@
+diffusers==0.25.1
+torch==2.0.0
+torchvision==0.15.1
+transformers==4.37.1
+accelerate
+safetensors
+einops
+onnxruntime-gpu
+spaces==0.19.4
+omegaconf
+peft
+huggingface-hub==0.20.2
+opencv-python
+insightface
+gradio
+controlnet_aux
+gdown
+peft
+spaces

style_template.py

@@ -0,0 +1,54 @@
+style_list = [
+    {
+        "name": "(No style)",
+        "prompt": "{prompt}",
+        "negative_prompt": "",
+    },
+    {
+        "name": "Spring Festival",
+        "prompt": "Flat illustration, a Chinese {prompt}, ancient style, wearing a red cloth, smile face, white skin, clean background, fireworks blooming, red lanterns",
+        "negative_prompt": "photo, deformed, black and white, realism, disfigured, low contrast, realistic, cropped, worst quality, missing fingers, extra digit, jpeg artifacts, signature, multiple, (lowres, low quality, worst quality:1.2)",
+    },
+    {
+        "name": "Watercolor",
+        "prompt": "watercolor painting, {prompt}. vibrant, beautiful, painterly, detailed, textural, artistic",
+        "negative_prompt": "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, anime, photorealistic, 35mm film, deformed, glitch, low contrast, noisy",
+    },
+    {
+        "name": "Film Noir",
+        "prompt": "film noir style, ink sketch|vector, {prompt} highly detailed, sharp focus, ultra sharpness, monochrome, high contrast, dramatic shadows, 1940s style, mysterious, cinematic",
+        "negative_prompt": "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
+    },
+    {
+        "name": "Neon",
+        "prompt": "masterpiece painting, buildings in the backdrop, kaleidoscope, lilac orange blue cream fuchsia bright vivid gradient colors, the scene is cinematic, {prompt}, emotional realism, double exposure, watercolor ink pencil, graded wash, color layering, magic realism, figurative painting, intricate motifs, organic tracery, polished",
+        "negative_prompt": "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
+    },
+    {
+        "name": "Jungle",
+        "prompt": 'waist-up "{prompt} in a Jungle" by Syd Mead, tangerine cold color palette, muted colors, detailed, 8k,photo r3al,dripping paint,3d toon style,3d style,Movie Still',
+        "negative_prompt": "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
+    },
+    {
+        "name": "Mars",
+        "prompt": "{prompt}, Post-apocalyptic. Mars Colony, Scavengers roam the wastelands searching for valuable resources, rovers, bright morning sunlight shining, (detailed) (intricate) (8k) (HDR) (cinematic lighting) (sharp focus)",
+        "negative_prompt": "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
+    },
+    {
+        "name": "Vibrant Color",
+        "prompt": "vibrant colorful, ink sketch|vector|2d colors, at nightfall, sharp focus, {prompt}, highly detailed, sharp focus, the clouds,colorful,ultra sharpness",
+        "negative_prompt": "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
+    },
+    {
+        "name": "Snow",
+        "prompt": "cinema 4d render, {prompt}, high contrast, vibrant and saturated, sico style, surrounded by magical glow,floating ice shards, snow crystals, cold, windy background, frozen natural landscape in background  cinematic atmosphere,highly detailed, sharp focus, intricate design, 3d, unreal engine, octane render, CG best quality, highres, photorealistic, dramatic lighting, artstation, concept art, cinematic, epic Steven Spielberg movie still, sharp focus, smoke, sparks, art by pascal blanche and greg rutkowski and repin, trending on artstation, hyperrealism painting, matte painting, 4k resolution",
+        "negative_prompt": "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, photo, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green",
+    },
+    {
+        "name": "Line art",
+        "prompt": "line art drawing {prompt} . professional, sleek, modern, minimalist, graphic, line art, vector graphics",
+        "negative_prompt": "anime, photorealistic, 35mm film, deformed, glitch, blurry, noisy, off-center, deformed, cross-eyed, closed eyes, bad anatomy, ugly, disfigured, mutated, realism, realistic, impressionism, expressionism, oil, acrylic",
+    },
+]
+
+styles = {k["name"]: (k["prompt"], k["negative_prompt"]) for k in style_list}

torchhub/README.md

@@ -0,0 +1,3 @@
+# Local PyTorch Hub
+
+This directory is for loading the DINOv2 encoder locally in case of no Internet connection.

torchhub/facebookresearch_dinov2_main/CODE_OF_CONDUCT.md

@@ -0,0 +1,80 @@
+# Code of Conduct
+
+## Our Pledge
+
+In the interest of fostering an open and welcoming environment, we as
+contributors and maintainers pledge to make participation in our project and
+our community a harassment-free experience for everyone, regardless of age, body
+size, disability, ethnicity, sex characteristics, gender identity and expression,
+level of experience, education, socio-economic status, nationality, personal
+appearance, race, religion, or sexual identity and orientation.
+
+## Our Standards
+
+Examples of behavior that contributes to creating a positive environment
+include:
+
+* Using welcoming and inclusive language
+* Being respectful of differing viewpoints and experiences
+* Gracefully accepting constructive criticism
+* Focusing on what is best for the community
+* Showing empathy towards other community members
+
+Examples of unacceptable behavior by participants include:
+
+* The use of sexualized language or imagery and unwelcome sexual attention or
+advances
+* Trolling, insulting/derogatory comments, and personal or political attacks
+* Public or private harassment
+* Publishing others' private information, such as a physical or electronic
+address, without explicit permission
+* Other conduct which could reasonably be considered inappropriate in a
+professional setting
+
+## Our Responsibilities
+
+Project maintainers are responsible for clarifying the standards of acceptable
+behavior and are expected to take appropriate and fair corrective action in
+response to any instances of unacceptable behavior.
+
+Project maintainers have the right and responsibility to remove, edit, or
+reject comments, commits, code, wiki edits, issues, and other contributions
+that are not aligned to this Code of Conduct, or to ban temporarily or
+permanently any contributor for other behaviors that they deem inappropriate,
+threatening, offensive, or harmful.
+
+## Scope
+
+This Code of Conduct applies within all project spaces, and it also applies when
+an individual is representing the project or its community in public spaces.
+Examples of representing a project or community include using an official
+project e-mail address, posting via an official social media account, or acting
+as an appointed representative at an online or offline event. Representation of
+a project may be further defined and clarified by project maintainers.
+
+This Code of Conduct also applies outside the project spaces when there is a
+reasonable belief that an individual's behavior may have a negative impact on
+the project or its community.
+
+## Enforcement
+
+Instances of abusive, harassing, or otherwise unacceptable behavior may be
+reported by contacting the project team at <opensource-conduct@meta.com>. All
+complaints will be reviewed and investigated and will result in a response that
+is deemed necessary and appropriate to the circumstances. The project team is
+obligated to maintain confidentiality with regard to the reporter of an incident.
+Further details of specific enforcement policies may be posted separately.
+
+Project maintainers who do not follow or enforce the Code of Conduct in good
+faith may face temporary or permanent repercussions as determined by other
+members of the project's leadership.
+
+## Attribution
+
+This Code of Conduct is adapted from the [Contributor Covenant][homepage], version 1.4,
+available at https://www.contributor-covenant.org/version/1/4/code-of-conduct.html
+
+[homepage]: https://www.contributor-covenant.org
+
+For answers to common questions about this code of conduct, see
+https://www.contributor-covenant.org/faq

torchhub/facebookresearch_dinov2_main/CONTRIBUTING.md

@@ -0,0 +1,31 @@
+# Contributing to DINOv2
+We want to make contributing to this project as easy and transparent as
+possible.
+
+## Pull Requests
+We actively welcome your pull requests.
+
+1. Fork the repo and create your branch from `main`.
+2. If you've added code that should be tested, add tests.
+3. If you've changed APIs, update the documentation.
+4. Ensure the test suite passes.
+5. Make sure your code lints.
+6. If you haven't already, complete the Contributor License Agreement ("CLA").
+
+## Contributor License Agreement ("CLA")
+In order to accept your pull request, we need you to submit a CLA. You only need
+to do this once to work on any of Meta's open source projects.
+
+Complete your CLA here: <https://code.facebook.com/cla>
+
+## Issues
+We use GitHub issues to track public bugs. Please ensure your description is
+clear and has sufficient instructions to be able to reproduce the issue.
+
+Meta has a [bounty program](https://www.facebook.com/whitehat/) for the safe
+disclosure of security bugs. In those cases, please go through the process
+outlined on that page and do not file a public issue.
+
+## License
+By contributing to DINOv2, you agree that your contributions will be licensed
+under the LICENSE file in the root directory of this source tree.

torchhub/facebookresearch_dinov2_main/LICENSE

@@ -0,0 +1,400 @@
+
+Attribution-NonCommercial 4.0 International
+
+=======================================================================
+
+Creative Commons Corporation ("Creative Commons") is not a law firm and
+does not provide legal services or legal advice. Distribution of
+Creative Commons public licenses does not create a lawyer-client or
+other relationship. Creative Commons makes its licenses and related
+information available on an "as-is" basis. Creative Commons gives no
+warranties regarding its licenses, any material licensed under their
+terms and conditions, or any related information. Creative Commons
+disclaims all liability for damages resulting from their use to the
+fullest extent possible.
+
+Using Creative Commons Public Licenses
+
+Creative Commons public licenses provide a standard set of terms and
+conditions that creators and other rights holders may use to share
+original works of authorship and other material subject to copyright
+and certain other rights specified in the public license below. The
+following considerations are for informational purposes only, are not
+exhaustive, and do not form part of our licenses.
+
+     Considerations for licensors: Our public licenses are
+     intended for use by those authorized to give the public
+     permission to use material in ways otherwise restricted by
+     copyright and certain other rights. Our licenses are
+     irrevocable. Licensors should read and understand the terms
+     and conditions of the license they choose before applying it.
+     Licensors should also secure all rights necessary before
+     applying our licenses so that the public can reuse the
+     material as expected. Licensors should clearly mark any
+     material not subject to the license. This includes other CC-
+     licensed material, or material used under an exception or
+     limitation to copyright. More considerations for licensors:
+	wiki.creativecommons.org/Considerations_for_licensors
+
+     Considerations for the public: By using one of our public
+     licenses, a licensor grants the public permission to use the
+     licensed material under specified terms and conditions. If
+     the licensor's permission is not necessary for any reason--for
+     example, because of any applicable exception or limitation to
+     copyright--then that use is not regulated by the license. Our
+     licenses grant only permissions under copyright and certain
+     other rights that a licensor has authority to grant. Use of
+     the licensed material may still be restricted for other
+     reasons, including because others have copyright or other
+     rights in the material. A licensor may make special requests,
+     such as asking that all changes be marked or described.
+     Although not required by our licenses, you are encouraged to
+     respect those requests where reasonable. More_considerations
+     for the public: 
+	wiki.creativecommons.org/Considerations_for_licensees
+
+=======================================================================
+
+Creative Commons Attribution-NonCommercial 4.0 International Public
+License
+
+By exercising the Licensed Rights (defined below), You accept and agree
+to be bound by the terms and conditions of this Creative Commons
+Attribution-NonCommercial 4.0 International Public License ("Public
+License"). To the extent this Public License may be interpreted as a
+contract, You are granted the Licensed Rights in consideration of Your
+acceptance of these terms and conditions, and the Licensor grants You
+such rights in consideration of benefits the Licensor receives from
+making the Licensed Material available under these terms and
+conditions.
+
+Section 1 -- Definitions.
+
+  a. Adapted Material means material subject to Copyright and Similar
+     Rights that is derived from or based upon the Licensed Material
+     and in which the Licensed Material is translated, altered,
+     arranged, transformed, or otherwise modified in a manner requiring
+     permission under the Copyright and Similar Rights held by the
+     Licensor. For purposes of this Public License, where the Licensed
+     Material is a musical work, performance, or sound recording,
+     Adapted Material is always produced where the Licensed Material is
+     synched in timed relation with a moving image.
+
+  b. Adapter's License means the license You apply to Your Copyright
+     and Similar Rights in Your contributions to Adapted Material in
+     accordance with the terms and conditions of this Public License.
+
+  c. Copyright and Similar Rights means copyright and/or similar rights
+     closely related to copyright including, without limitation,
+     performance, broadcast, sound recording, and Sui Generis Database
+     Rights, without regard to how the rights are labeled or
+     categorized. For purposes of this Public License, the rights
+     specified in Section 2(b)(1)-(2) are not Copyright and Similar
+     Rights.
+  d. Effective Technological Measures means those measures that, in the
+     absence of proper authority, may not be circumvented under laws
+     fulfilling obligations under Article 11 of the WIPO Copyright
+     Treaty adopted on December 20, 1996, and/or similar international
+     agreements.
+
+  e. Exceptions and Limitations means fair use, fair dealing, and/or
+     any other exception or limitation to Copyright and Similar Rights
+     that applies to Your use of the Licensed Material.
+
+  f. Licensed Material means the artistic or literary work, database,
+     or other material to which the Licensor applied this Public
+     License.
+
+  g. Licensed Rights means the rights granted to You subject to the
+     terms and conditions of this Public License, which are limited to
+     all Copyright and Similar Rights that apply to Your use of the
+     Licensed Material and that the Licensor has authority to license.
+
+  h. Licensor means the individual(s) or entity(ies) granting rights
+     under this Public License.
+
+  i. NonCommercial means not primarily intended for or directed towards
+     commercial advantage or monetary compensation. For purposes of
+     this Public License, the exchange of the Licensed Material for
+     other material subject to Copyright and Similar Rights by digital
+     file-sharing or similar means is NonCommercial provided there is
+     no payment of monetary compensation in connection with the
+     exchange.
+
+  j. Share means to provide material to the public by any means or
+     process that requires permission under the Licensed Rights, such
+     as reproduction, public display, public performance, distribution,
+     dissemination, communication, or importation, and to make material
+     available to the public including in ways that members of the
+     public may access the material from a place and at a time
+     individually chosen by them.
+
+  k. Sui Generis Database Rights means rights other than copyright
+     resulting from Directive 96/9/EC of the European Parliament and of
+     the Council of 11 March 1996 on the legal protection of databases,
+     as amended and/or succeeded, as well as other essentially
+     equivalent rights anywhere in the world.
+
+  l. You means the individual or entity exercising the Licensed Rights
+     under this Public License. Your has a corresponding meaning.
+
+Section 2 -- Scope.
+
+  a. License grant.
+
+       1. Subject to the terms and conditions of this Public License,
+          the Licensor hereby grants You a worldwide, royalty-free,
+          non-sublicensable, non-exclusive, irrevocable license to
+          exercise the Licensed Rights in the Licensed Material to:
+
+            a. reproduce and Share the Licensed Material, in whole or
+               in part, for NonCommercial purposes only; and
+
+            b. produce, reproduce, and Share Adapted Material for
+               NonCommercial purposes only.
+
+       2. Exceptions and Limitations. For the avoidance of doubt, where
+          Exceptions and Limitations apply to Your use, this Public
+          License does not apply, and You do not need to comply with
+          its terms and conditions.
+
+       3. Term. The term of this Public License is specified in Section
+          6(a).
+
+       4. Media and formats; technical modifications allowed. The
+          Licensor authorizes You to exercise the Licensed Rights in
+          all media and formats whether now known or hereafter created,
+          and to make technical modifications necessary to do so. The
+          Licensor waives and/or agrees not to assert any right or
+          authority to forbid You from making technical modifications
+          necessary to exercise the Licensed Rights, including
+          technical modifications necessary to circumvent Effective
+          Technological Measures. For purposes of this Public License,
+          simply making modifications authorized by this Section 2(a)
+          (4) never produces Adapted Material.
+
+       5. Downstream recipients.
+
+            a. Offer from the Licensor -- Licensed Material. Every
+               recipient of the Licensed Material automatically
+               receives an offer from the Licensor to exercise the
+               Licensed Rights under the terms and conditions of this
+               Public License.
+
+            b. No downstream restrictions. You may not offer or impose
+               any additional or different terms or conditions on, or
+               apply any Effective Technological Measures to, the
+               Licensed Material if doing so restricts exercise of the
+               Licensed Rights by any recipient of the Licensed
+               Material.
+
+       6. No endorsement. Nothing in this Public License constitutes or
+          may be construed as permission to assert or imply that You
+          are, or that Your use of the Licensed Material is, connected
+          with, or sponsored, endorsed, or granted official status by,
+          the Licensor or others designated to receive attribution as
+          provided in Section 3(a)(1)(A)(i).
+
+  b. Other rights.
+
+       1. Moral rights, such as the right of integrity, are not
+          licensed under this Public License, nor are publicity,
+          privacy, and/or other similar personality rights; however, to
+          the extent possible, the Licensor waives and/or agrees not to
+          assert any such rights held by the Licensor to the limited
+          extent necessary to allow You to exercise the Licensed
+          Rights, but not otherwise.
+
+       2. Patent and trademark rights are not licensed under this
+          Public License.
+
+       3. To the extent possible, the Licensor waives any right to
+          collect royalties from You for the exercise of the Licensed
+          Rights, whether directly or through a collecting society
+          under any voluntary or waivable statutory or compulsory
+          licensing scheme. In all other cases the Licensor expressly
+          reserves any right to collect such royalties, including when
+          the Licensed Material is used other than for NonCommercial
+          purposes.
+
+Section 3 -- License Conditions.
+
+Your exercise of the Licensed Rights is expressly made subject to the
+following conditions.
+
+  a. Attribution.
+
+       1. If You Share the Licensed Material (including in modified
+          form), You must:
+
+            a. retain the following if it is supplied by the Licensor
+               with the Licensed Material:
+
+                 i. identification of the creator(s) of the Licensed
+                    Material and any others designated to receive
+                    attribution, in any reasonable manner requested by
+                    the Licensor (including by pseudonym if
+                    designated);
+
+                ii. a copyright notice;
+
+               iii. a notice that refers to this Public License;
+
+                iv. a notice that refers to the disclaimer of
+                    warranties;
+
+                 v. a URI or hyperlink to the Licensed Material to the
+                    extent reasonably practicable;
+
+            b. indicate if You modified the Licensed Material and
+               retain an indication of any previous modifications; and
+
+            c. indicate the Licensed Material is licensed under this
+               Public License, and include the text of, or the URI or
+               hyperlink to, this Public License.
+
+       2. You may satisfy the conditions in Section 3(a)(1) in any
+          reasonable manner based on the medium, means, and context in
+          which You Share the Licensed Material. For example, it may be
+          reasonable to satisfy the conditions by providing a URI or
+          hyperlink to a resource that includes the required
+          information.
+
+       3. If requested by the Licensor, You must remove any of the
+          information required by Section 3(a)(1)(A) to the extent
+          reasonably practicable.
+
+       4. If You Share Adapted Material You produce, the Adapter's
+          License You apply must not prevent recipients of the Adapted
+          Material from complying with this Public License.
+
+Section 4 -- Sui Generis Database Rights.
+
+Where the Licensed Rights include Sui Generis Database Rights that
+apply to Your use of the Licensed Material:
+
+  a. for the avoidance of doubt, Section 2(a)(1) grants You the right
+     to extract, reuse, reproduce, and Share all or a substantial
+     portion of the contents of the database for NonCommercial purposes
+     only;
+
+  b. if You include all or a substantial portion of the database
+     contents in a database in which You have Sui Generis Database
+     Rights, then the database in which You have Sui Generis Database
+     Rights (but not its individual contents) is Adapted Material; and
+
+  c. You must comply with the conditions in Section 3(a) if You Share
+     all or a substantial portion of the contents of the database.
+
+For the avoidance of doubt, this Section 4 supplements and does not
+replace Your obligations under this Public License where the Licensed
+Rights include other Copyright and Similar Rights.
+
+Section 5 -- Disclaimer of Warranties and Limitation of Liability.
+
+  a. UNLESS OTHERWISE SEPARATELY UNDERTAKEN BY THE LICENSOR, TO THE
+     EXTENT POSSIBLE, THE LICENSOR OFFERS THE LICENSED MATERIAL AS-IS
+     AND AS-AVAILABLE, AND MAKES NO REPRESENTATIONS OR WARRANTIES OF
+     ANY KIND CONCERNING THE LICENSED MATERIAL, WHETHER EXPRESS,
+     IMPLIED, STATUTORY, OR OTHER. THIS INCLUDES, WITHOUT LIMITATION,
+     WARRANTIES OF TITLE, MERCHANTABILITY, FITNESS FOR A PARTICULAR
+     PURPOSE, NON-INFRINGEMENT, ABSENCE OF LATENT OR OTHER DEFECTS,
+     ACCURACY, OR THE PRESENCE OR ABSENCE OF ERRORS, WHETHER OR NOT
+     KNOWN OR DISCOVERABLE. WHERE DISCLAIMERS OF WARRANTIES ARE NOT
+     ALLOWED IN FULL OR IN PART, THIS DISCLAIMER MAY NOT APPLY TO YOU.
+
+  b. TO THE EXTENT POSSIBLE, IN NO EVENT WILL THE LICENSOR BE LIABLE
+     TO YOU ON ANY LEGAL THEORY (INCLUDING, WITHOUT LIMITATION,
+     NEGLIGENCE) OR OTHERWISE FOR ANY DIRECT, SPECIAL, INDIRECT,
+     INCIDENTAL, CONSEQUENTIAL, PUNITIVE, EXEMPLARY, OR OTHER LOSSES,
+     COSTS, EXPENSES, OR DAMAGES ARISING OUT OF THIS PUBLIC LICENSE OR
+     USE OF THE LICENSED MATERIAL, EVEN IF THE LICENSOR HAS BEEN
+     ADVISED OF THE POSSIBILITY OF SUCH LOSSES, COSTS, EXPENSES, OR
+     DAMAGES. WHERE A LIMITATION OF LIABILITY IS NOT ALLOWED IN FULL OR
+     IN PART, THIS LIMITATION MAY NOT APPLY TO YOU.
+
+  c. The disclaimer of warranties and limitation of liability provided
+     above shall be interpreted in a manner that, to the extent
+     possible, most closely approximates an absolute disclaimer and
+     waiver of all liability.
+
+Section 6 -- Term and Termination.
+
+  a. This Public License applies for the term of the Copyright and
+     Similar Rights licensed here. However, if You fail to comply with
+     this Public License, then Your rights under this Public License
+     terminate automatically.
+
+  b. Where Your right to use the Licensed Material has terminated under
+     Section 6(a), it reinstates:
+
+       1. automatically as of the date the violation is cured, provided
+          it is cured within 30 days of Your discovery of the
+          violation; or
+
+       2. upon express reinstatement by the Licensor.
+
+     For the avoidance of doubt, this Section 6(b) does not affect any
+     right the Licensor may have to seek remedies for Your violations
+     of this Public License.
+
+  c. For the avoidance of doubt, the Licensor may also offer the
+     Licensed Material under separate terms or conditions or stop
+     distributing the Licensed Material at any time; however, doing so
+     will not terminate this Public License.
+
+  d. Sections 1, 5, 6, 7, and 8 survive termination of this Public
+     License.
+
+Section 7 -- Other Terms and Conditions.
+
+  a. The Licensor shall not be bound by any additional or different
+     terms or conditions communicated by You unless expressly agreed.
+
+  b. Any arrangements, understandings, or agreements regarding the
+     Licensed Material not stated herein are separate from and
+     independent of the terms and conditions of this Public License.
+
+Section 8 -- Interpretation.
+
+  a. For the avoidance of doubt, this Public License does not, and
+     shall not be interpreted to, reduce, limit, restrict, or impose
+     conditions on any use of the Licensed Material that could lawfully
+     be made without permission under this Public License.
+
+  b. To the extent possible, if any provision of this Public License is
+     deemed unenforceable, it shall be automatically reformed to the
+     minimum extent necessary to make it enforceable. If the provision
+     cannot be reformed, it shall be severed from this Public License
+     without affecting the enforceability of the remaining terms and
+     conditions.
+
+  c. No term or condition of this Public License will be waived and no
+     failure to comply consented to unless expressly agreed to by the
+     Licensor.
+
+  d. Nothing in this Public License constitutes or may be interpreted
+     as a limitation upon, or waiver of, any privileges and immunities
+     that apply to the Licensor or You, including from the legal
+     processes of any jurisdiction or authority.
+
+=======================================================================
+
+Creative Commons is not a party to its public
+licenses. Notwithstanding, Creative Commons may elect to apply one of
+its public licenses to material it publishes and in those instances
+will be considered the “Licensor.” The text of the Creative Commons
+public licenses is dedicated to the public domain under the CC0 Public
+Domain Dedication. Except for the limited purpose of indicating that
+material is shared under a Creative Commons public license or as
+otherwise permitted by the Creative Commons policies published at
+creativecommons.org/policies, Creative Commons does not authorize the
+use of the trademark "Creative Commons" or any other trademark or logo
+of Creative Commons without its prior written consent including,
+without limitation, in connection with any unauthorized modifications
+to any of its public licenses or any other arrangements,
+understandings, or agreements concerning use of licensed material. For
+the avoidance of doubt, this paragraph does not form part of the
+public licenses.
+
+Creative Commons may be contacted at creativecommons.org.

torchhub/facebookresearch_dinov2_main/MODEL_CARD.md

@@ -0,0 +1,201 @@
+# Model Card for DINOv2-S/B/L/g
+
+These are Vision Transformer models trained following the method described in the paper:
+"DINOv2: Learning Robust Visual Features without Supervision"
+
+We provide 4 models: 1 ViT-g trained from scratch, and 3 ViT-S/B/L models distilled from the ViT-g.
+
+## Model Details
+The model takes an image as input and returns a class token and patch tokens.
+
+The embedding dimension is: 
+- 384 for ViT-S.
+- 768 for ViT-B.
+- 1024 for ViT-L.
+- 1536 for ViT-g.
+
+The models follow a Transformer architecture, with a patch size of 14.
+
+For a 224x224 image, this results in 1 class token + 256 patch tokens.
+
+The models can accept larger images provided the image shapes are multiples of the patch size (14). 
+If this condition is not verified, the model will crop to the closest smaller multiple of the patch size.
+
+### Model Description
+
+- **Developed by:** Meta AI
+- **Model type:** Vision Transformer
+- **License:** CC-BY-NC
+
+- **Repository:** https://github.com/facebookresearch/dinov2
+- **Paper:** https://arxiv.org/abs/2304.07193
+- **Demo:** https://dinov2.metademolab.com/
+
+## Uses
+
+The models are vision backbones providing multi-purpose features for downstream tasks.
+
+### Direct Use
+
+The models can be used without fine-tuning, with downstream classifiers as simple as linear layers, to obtain competitive results:
+- on depth estimation, semantic segmentation, using linear layers.
+- on image classification, using k-NN classifiers on the class token.
+- on image classification, with logistic regression classifiers applied on the class token.
+- on image classification, with a linear layer applied on the class token and the average of the patch tokens.
+- on image retrieval using nearest neighbors.
+
+### Downstream Use
+
+It is technically possible to perform fine-tuning on the models, for small gains (we measured +2% on ImageNet-1k classification). 
+We recommend keeping this as a very last step and only when necessary, as the features already provide good performance out-of-the-box.
+
+## Bias, Risks, and Limitations
+
+Despite improvements thanks to the training method not using annotations, we still observe significant biases in our models toward rich households from Western countries.
+
+### Recommendations
+
+We expect fine-tuning will increase the biases in the features produced by the model as they will be tuned to the fine-tuning labels.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+```python
+import torch
+dinov2_vits14 = torch.hub.load('facebookresearch/dinov2', 'dinov2_vits14')
+dinov2_vitb14 = torch.hub.load('facebookresearch/dinov2', 'dinov2_vitb14')
+dinov2_vitl14 = torch.hub.load('facebookresearch/dinov2', 'dinov2_vitl14')
+dinov2_vitg14 = torch.hub.load('facebookresearch/dinov2', 'dinov2_vitg14')
+```
+
+## Training Details
+
+### Training Data
+
+- **Training data:** LVD-142M (see paper)
+- **Training regime:** fp16 using PyTorch-FSDP mixed-precision.
+
+### Training Procedure 
+
+- **Training objective:**
+  - DINO self-distillation loss with multi-crop
+  - iBOT masked-image modeling loss
+  - KoLeo regularization on [CLS] tokens
+- **Architectures:**
+  - ViT-S (21M params): Patch size 14, embedding dimension 384, 6 heads, MLP FFN
+  - ViT-B (86M params): Patch size 14, embedding dimension 768, 12 heads, MLP FFN
+  - ViT-L (0.3B params): Patch size 14, embedding dimension 1024, 16 heads, MLP FFN
+  - ViT-g (1.1B params): Patch size 14, embedding dimension 1536, 24 heads, SwiGLU FFN
+- **Distillation:**
+  - Distillation follows the standard DINOv2 pretraining procedure, except the teacher is a pretrained ViT-g, frozen.
+
+## Evaluation
+
+We refer users to the associated paper for the evaluation protocols.
+
+<table>
+  <tr>
+    <th>model</th>
+    <th colspan="3">ImageNet-1k</th>
+    <th>NYU-Depth v2</th>
+    <th>SUN-RGBD</th>
+    <th>ADE20k</th>
+    <th>iNaturalist 2018</th>
+    <th>Oxford-H</th>
+  </tr>
+  <tr>
+    <th rowspan="2">task</th>
+    <th>classif. (acc)</th>
+    <th>classif. (acc)</th>
+    <th>classif. V2 (acc)</th>
+    <th>depth (RMSE)</th>
+    <th>depth (RMSE)</th>
+    <th>segm. (mAP)</th>
+    <th>classif. (acc)</th>
+    <th>retrieval (mAP)</th>
+  </tr>
+  <tr>
+    <!-- <th>^</th> -->
+    <th>k-NN</th>
+    <th>linear</th>
+    <th>linear</th>
+    <th>linear<br />4 layers</th>
+    <th>NYU-D transfer</th>
+    <th>multiscale</th>
+    <th>linear</th>
+    <th>nearest neighbor</th>
+  </tr>
+  <tr>
+    <td>ViT-S/14</td>
+    <td align="right">79.0%</td>
+    <td align="right">81.1%</td>
+    <td align="right">70.8%</td> 
+    <td align="right">0.417</td> 
+    <td align="right">0.431</td> 
+    <td align="right">47.2</td> 
+    <td align="right">69.5%</td> 
+    <td align="right">43.2</td> 
+  </tr>
+  <tr>
+    <td>ViT-B/14</td>
+    <td align="right">82.1%</td>
+    <td align="right">84.5%</td>
+    <td align="right">74.9%</td>
+    <td align="right">0.362</td> 
+    <td align="right">0.400</td> 
+    <td align="right">51.3</td> 
+    <td align="right">76.3%</td> 
+    <td align="right">49.5</td> 
+  </tr>
+  <tr>
+    <td>ViT-L/14</td>
+    <td align="right">83.5%</td>
+    <td align="right">86.3%</td>
+    <td align="right">77.6%</td>
+    <td align="right">0.333</td> 
+    <td align="right">0.396</td> 
+    <td align="right">53.1</td> 
+    <td align="right">79.8%</td> 
+    <td align="right">54.0</td> 
+  </tr>
+  <tr>
+    <td>ViT-g/14</td>
+    <td align="right">83.5%</td>
+    <td align="right">86.5%</td>
+    <td align="right">78.4%</td>
+    <td align="right">0.298</td> 
+    <td align="right">0.362</td> 
+    <td align="right">53.0</td> 
+    <td align="right">81.6%</td> 
+    <td align="right">52.3</td> 
+  </tr>
+</table>
+
+## Environmental Impact
+
+- **Hardware Type:** Nvidia A100
+- **Hours used:** 22,000 for ViT-g, 4,500 for ViT-S distillation, 5,300 for ViT-B distillation, 8,000 for ViT-L distillation
+- **Cloud Provider:** Private infra
+- **Compute Region:** USA
+- **Carbon Emitted:** 7t CO2eq
+
+#### Hardware
+
+Nvidia A100 GPUs
+
+#### Software
+
+PyTorch 2.0,
+xFormers 0.0.18
+
+**BibTeX**
+
+```
+@misc{oquab2023dinov2,
+  title={DINOv2: Learning Robust Visual Features without Supervision},
+  author={Oquab, Maxime and Darcet, Timothée and Moutakanni, Theo and Vo, Huy and Szafraniec, Marc and Khalidov, Vasil and Fernandez, Pierre and Haziza, Daniel and Massa, Francisco and El-Nouby, Alaaeldin and Howes, Russell and Huang, Po-Yao and Xu, Hu and Sharma, Vasu and Li, Shang-Wen and Galuba, Wojciech and Rabbat, Mike and Assran, Mido and Ballas, Nicolas and Synnaeve, Gabriel and Misra, Ishan and Jegou, Herve and Mairal, Julien and Labatut, Patrick and Joulin, Armand and Bojanowski, Piotr},
+  journal={arXiv:2304.07193},
+  year={2023}
+}
+```

torchhub/facebookresearch_dinov2_main/README.md

@@ -0,0 +1,277 @@
+# DINOv2: Learning Robust Visual Features without Supervision
+
+**[Meta AI Research, FAIR](https://ai.facebook.com/research/)**
+
+Maxime Oquab,
+Timothée Darcet,
+Théo Moutakanni,
+Huy V. Vo,
+Marc Szafraniec,
+Vasil Khalidov,
+Patrick Labatut,
+Armand Joulin,
+Piotr Bojanowski
+
+[[`Paper`](https://arxiv.org/abs/2304.07193)] [[`Blog`](https://ai.facebook.com/blog/dino-v2-computer-vision-self-supervised-learning/)] [[`Demo`](https://dinov2.metademolab.com)] [[`BibTeX`](#citing-dinov2)]
+
+PyTorch implementation and pretrained models for DINOv2. For details, see the paper: **[DINOv2: Learning Robust Visual Features without Supervision](https://arxiv.org/abs/2304.07193)**.
+
+DINOv2 models produce high-performance visual features that can be directly employed with classifiers as simple as linear layers on a variety of computer vision tasks; these visual features are robust and perform well across domains without any requirement for fine-tuning. The models were pretrained on a dataset of 142 M images without using any labels or annotations.
+
+https://github.com/facebookresearch/dinov2/assets/60359573/f168823e-7922-415a-b429-578badf5c356
+
+<div align="center">
+  Visualization of the three first principal components of the patch features of all frames, mapped to RGB values.
+</div>
+
+## Pretrained models
+
+<table style="margin: auto">
+  <tr>
+    <th>model</th>
+    <th># of<br />params</th>
+    <th>ImageNet<br />k-NN</th>
+    <th>ImageNet<br />linear</th>
+    <th>download</th>
+  </tr>
+  <tr>
+    <td>ViT-S/14 distilled</td>
+    <td align="right">21 M</td>
+    <td align="right">79.0%</td>
+    <td align="right">81.1%</td>
+    <td><a href="https://dl.fbaipublicfiles.com/dinov2/dinov2_vits14/dinov2_vits14_pretrain.pth">backbone only</a></td>
+  </tr>
+  <tr>
+    <td>ViT-B/14 distilled</td>
+    <td align="right">86 M</td>
+    <td align="right">82.1%</td>
+    <td align="right">84.5%</td>
+    <td><a href="https://dl.fbaipublicfiles.com/dinov2/dinov2_vitb14/dinov2_vitb14_pretrain.pth">backbone only</a></td>
+  </tr>
+  <tr>
+    <td>ViT-L/14 distilled</td>
+    <td align="right">300 M</td>
+    <td align="right">83.5%</td>
+    <td align="right">86.3%</td>
+    <td><a href="https://dl.fbaipublicfiles.com/dinov2/dinov2_vitl14/dinov2_vitl14_pretrain.pth">backbone only</a></td>
+  </tr>
+  <tr>
+    <td>ViT-g/14</td>
+    <td align="right">1,100 M</td>
+    <td align="right">83.5%</td>
+    <td align="right">86.5%</td>
+    <td><a href="https://dl.fbaipublicfiles.com/dinov2/dinov2_vitg14/dinov2_vitg14_pretrain.pth">backbone only</a></td>
+  </tr>
+</table>
+
+### Pretrained models via PyTorch Hub
+
+Please follow the instructions [here](https://pytorch.org/get-started/locally/) to install PyTorch (the only required dependency for loading the model). Installing PyTorch with CUDA support is strongly recommended.
+
+A corresponding [model card](MODEL_CARD.md) is included in the repository.
+
+```python
+import torch
+
+dinov2_vits14 = torch.hub.load('facebookresearch/dinov2', 'dinov2_vits14')
+dinov2_vitb14 = torch.hub.load('facebookresearch/dinov2', 'dinov2_vitb14')
+dinov2_vitl14 = torch.hub.load('facebookresearch/dinov2', 'dinov2_vitl14')
+dinov2_vitg14 = torch.hub.load('facebookresearch/dinov2', 'dinov2_vitg14')
+```
+
+## Installation
+
+The training and evaluation code requires PyTorch 2.0 and [xFormers](https://github.com/facebookresearch/xformers) 0.0.18 as well as a number of other 3rd party packages. Note that the code has only been tested with the specified versions and also expects a Linux environment. To setup all the required dependencies for training and evaluation, please follow the instructions below:
+
+*[conda](https://docs.conda.io/projects/conda/en/latest/user-guide/getting-started.html)* **(Recommended)** - Clone the repository and then create and activate a `dinov2` conda environment using the provided environment definition:
+
+```shell
+conda env create -f conda.yaml
+conda activate dinov2
+```
+
+*[pip](https://pip.pypa.io/en/stable/getting-started/)* - Clone the repository and then use the provided `requirements.txt` to install the dependencies:
+
+```shell
+pip install -r requirements.txt
+```
+
+## Data preparation
+
+### ImageNet-1k
+
+The root directory of the dataset should hold the following contents:
+
+- `<ROOT>/test/ILSVRC2012_test_00000001.JPEG`
+- `<ROOT>/test/[..]`
+- `<ROOT>/test/ILSVRC2012_test_00100000.JPEG`
+- `<ROOT>/train/n01440764/n01440764_10026.JPEG`
+- `<ROOT>/train/[...]`
+- `<ROOT>/train/n15075141/n15075141_9993.JPEG`
+- `<ROOT>/val/n01440764/ILSVRC2012_val_00000293.JPEG`
+- `<ROOT>/val/[...]`
+- `<ROOT>/val/n15075141/ILSVRC2012_val_00049174.JPEG`
+- `<ROOT>/labels.txt`
+
+The provided dataset implementation expects a few additional metadata files to be present under the extra directory:
+
+- `<EXTRA>/class-ids-TRAIN.npy`
+- `<EXTRA>/class-ids-VAL.npy`
+- `<EXTRA>/class-names-TRAIN.npy`
+- `<EXTRA>/class-names-VAL.npy`
+- `<EXTRA>/entries-TEST.npy`
+- `<EXTRA>/entries-TRAIN.npy`
+- `<EXTRA>/entries-VAL.npy`
+
+These metadata files can be generated (once) with the following lines of Python code:
+
+```python
+from dinov2.data.datasets import ImageNet
+
+for split in ImageNet.Split:
+    dataset = ImageNet(split=split, root="<ROOT>", extra="<EXTRA>")
+    dataset.dump_extra()
+```
+
+Note that the root and extra directories do not have to be distinct directories.
+
+### ImageNet-22k
+
+Please adapt the [dataset class](dinov2/data/datasets/image_net_22k.py) to match your local setup.
+
+<br />
+
+:warning: To execute the commands provided in the next sections for training and evaluation, the `dinov2` package should be included in the Python module search path, i.e. simply prefix the command to run with `PYTHONPATH=.`.
+
+## Training
+
+### Fast setup: training DINOv2 ViT-L/16 on ImageNet-1k
+
+Run DINOv2 training on 4 A100-80GB nodes (32 GPUs) in a SLURM cluster environment with submitit:
+
+```shell
+python dinov2/run/train/train.py \
+    --nodes 4 \
+    --config-file dinov2/configs/train/vitl16_short.yaml \
+    --output-dir <PATH/TO/OUTPUT/DIR> \
+    train.dataset_path=ImageNet:split=TRAIN:root=<PATH/TO/DATASET>:extra=<PATH/TO/DATASET>
+```
+
+Training time is approximately 1 day and the resulting checkpoint should reach 81.6% on k-NN eval and 82.9% on linear eval.
+
+The training code saves the weights of the teacher in the `eval` folder every 12500 iterations for evaluation.
+
+### Long setup: training DINOv2 ViT-L/14 on ImageNet-22k
+
+Run DINOv2 training on 12 A100-80GB nodes (96 GPUs) in a SLURM cluster environment with submitit:
+
+```shell
+python dinov2/run/train/train.py \
+    --nodes 12 \
+    --config-file dinov2/configs/train/vitl14.yaml \
+    --output-dir <PATH/TO/OUTPUT/DIR> \
+    train.dataset_path=ImageNet22k:root=<PATH/TO/DATASET>:extra=<PATH/TO/DATASET>
+```
+
+Training time is approximately 3.3 days and the resulting checkpoint should reach 82.0% on k-NN eval and 84.5% on linear eval.
+
+The training code saves the weights of the teacher in the `eval` folder every 12500 iterations for evaluation.
+
+
+## Evaluation
+
+The training code regularly saves the teacher weights. In order to evaluate the model, run the following evaluation on a single node:
+
+### k-NN classification on ImageNet-1k
+
+```shell
+python dinov2/run/eval/knn.py \
+    --config-file <PATH/TO/OUTPUT/DIR>/config.yaml \
+    --pretrained-weights <PATH/TO/OUTPUT/DIR>/eval/training_24999/teacher_checkpoint.pth \
+    --output-dir <PATH/TO/OUTPUT/DIR>/eval/training_24999/knn \
+    --train-dataset ImageNet:split=TRAIN:root=<PATH/TO/DATASET>:extra=<PATH/TO/DATASET> \
+    --val-dataset ImageNet:split=VAL:root=<PATH/TO/DATASET>:extra=<PATH/TO/DATASET>
+```
+
+### Logistic regression classification on ImageNet-1k
+
+```shell
+python dinov2/run/eval/log_regression.py \
+    --config-file <PATH/TO/OUTPUT/DIR>/config.yaml \
+    --pretrained-weights <PATH/TO/OUTPUT/DIR>/eval/training_24999/teacher_checkpoint.pth \
+    --output-dir <PATH/TO/OUTPUT/DIR>/eval/training_24999/logreg \
+    --train-dataset ImageNet:split=TRAIN:root=<PATH/TO/DATASET>:extra=<PATH/TO/DATASET> \
+    --val-dataset ImageNet:split=VAL:root=<PATH/TO/DATASET>:extra=<PATH/TO/DATASET>
+```
+
+### Linear classification with data augmentation on ImageNet-1k
+
+```shell
+python dinov2/run/eval/linear.py \
+    --config-file <PATH/TO/OUTPUT/DIR>/config.yaml \
+    --pretrained-weights <PATH/TO/OUTPUT/DIR>/eval/training_24999/teacher_checkpoint.pth \
+    --output-dir <PATH/TO/OUTPUT/DIR>/eval/training_24999/linear \
+    --train-dataset ImageNet:split=TRAIN:root=<PATH/TO/DATASET>:extra=<PATH/TO/DATASET> \
+    --val-dataset ImageNet:split=VAL:root=<PATH/TO/DATASET>:extra=<PATH/TO/DATASET>
+```
+
+We release the weights from evaluating the different models:
+
+<table style="margin: auto">
+  <tr>
+    <th>model</th>
+    <th>ImageNet<br />top-1</th>
+    <th>linear evaluation</th>
+  </tr>
+  <tr>
+    <td>ViT-S/14 distilled</td>
+    <td align="right">81.1%</td>
+    <td><a href="https://dl.fbaipublicfiles.com/dinov2/dinov2_vits14/dinov2_vits14_linear_head.pth">linear head weights</a></td>
+  </tr>
+  <tr>
+    <td>ViT-B/14 distilled</td>
+    <td align="right">84.5%</td>
+    <td><a href="https://dl.fbaipublicfiles.com/dinov2/dinov2_vitb14/dinov2_vitb14_linear_head.pth">linear head weights</a></td>
+  </tr>
+  <tr>
+    <td>ViT-L/14 distilled</td>
+    <td align="right">86.3%</td>
+    <td><a href="https://dl.fbaipublicfiles.com/dinov2/dinov2_vitl14/dinov2_vitl14_linear_head.pth">linear head weights</a></td>
+  </tr>
+  <tr>
+    <td>ViT-g/14</td>
+    <td align="right">86.5%</td>
+    <td><a href="https://dl.fbaipublicfiles.com/dinov2/dinov2_vitg14/dinov2_vitg14_linear_head.pth">linear head weights</a></td>
+  </tr>
+</table>
+
+The performance of the provided pretrained model weights can be evaluated as follows on ImageNet-1k:
+
+```shell
+python dinov2/run/eval/linear.py \
+    --config-file dinov2/configs/eval/vitg14_pretrain.yaml \
+    --pretrained-weights https://dl.fbaipublicfiles.com/dinov2/dinov2_vitg14/dinov2_vitg14_pretrain.pth \
+    --train-dataset ImageNet:split=TRAIN:root=<PATH/TO/DATASET>:extra=<PATH/TO/DATASET> \
+    --val-dataset ImageNet:split=VAL:root=<PATH/TO/DATASET>:extra=<PATH/TO/DATASET>
+```
+
+## License
+
+DINOv2 code and model weights are released under the CC-BY-NC 4.0 license. See [LICENSE](LICENSE) for additional details.
+
+## Contributing
+
+See [contributing](CONTRIBUTING.md) and the [code of conduct](CODE_OF_CONDUCT.md).
+
+## Citing DINOv2
+
+If you find this repository useful, please consider giving a star :star: and citation :t-rex::
+
+```
+@misc{oquab2023dinov2,
+  title={DINOv2: Learning Robust Visual Features without Supervision},
+  author={Oquab, Maxime and Darcet, Timothée and Moutakanni, Theo and Vo, Huy V. and Szafraniec, Marc and Khalidov, Vasil and Fernandez, Pierre and Haziza, Daniel and Massa, Francisco and El-Nouby, Alaaeldin and Howes, Russell and Huang, Po-Yao and Xu, Hu and Sharma, Vasu and Li, Shang-Wen and Galuba, Wojciech and Rabbat, Mike and Assran, Mido and Ballas, Nicolas and Synnaeve, Gabriel and Misra, Ishan and Jegou, Herve and Mairal, Julien and Labatut, Patrick and Joulin, Armand and Bojanowski, Piotr},
+  journal={arXiv:2304.07193},
+  year={2023}
+}
+```

torchhub/facebookresearch_dinov2_main/conda.yaml

@@ -0,0 +1,22 @@
+name: dinov2
+channels:
+  - defaults
+  - pytorch
+  - nvidia
+  - xformers
+  - conda-forge
+dependencies:
+  - python=3.9
+  - pytorch::pytorch=2.0.0
+  - pytorch::pytorch-cuda=11.7.0
+  - pytorch::torchvision=0.15.0
+  - omegaconf
+  - torchmetrics=0.10.3
+  - fvcore
+  - iopath
+  - xformers::xformers=0.0.18
+  - pip
+  - pip:
+    - git+https://github.com/facebookincubator/submitit
+    - --extra-index-url https://pypi.nvidia.com
+    - cuml-cu11

torchhub/facebookresearch_dinov2_main/dinov2/__init__.py

@@ -0,0 +1,7 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+__version__ = "0.0.1"

torchhub/facebookresearch_dinov2_main/dinov2/configs/__init__.py

@@ -0,0 +1,23 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import pathlib
+
+from omegaconf import OmegaConf
+
+
+def load_config(config_name: str):
+    config_filename = config_name + ".yaml"
+    return OmegaConf.load(pathlib.Path(__file__).parent.resolve() / config_filename)
+
+
+dinov2_default_config = load_config("ssl_default_config")
+
+
+def load_and_merge_config(config_name: str):
+    default_config = OmegaConf.create(dinov2_default_config)
+    loaded_config = load_config(config_name)
+    return OmegaConf.merge(default_config, loaded_config)

torchhub/facebookresearch_dinov2_main/dinov2/configs/eval/vitb14_pretrain.yaml

@@ -0,0 +1,6 @@
+student:
+  arch: vit_base
+  patch_size: 14
+crops:
+  global_crops_size: 518  # this is to set up the position embeddings properly
+  local_crops_size: 98

torchhub/facebookresearch_dinov2_main/dinov2/configs/eval/vitg14_pretrain.yaml

@@ -0,0 +1,7 @@
+student:
+  arch: vit_giant2
+  patch_size: 14
+  ffn_layer: swiglufused
+crops:
+  global_crops_size: 518  # this is to set up the position embeddings properly
+  local_crops_size: 98

torchhub/facebookresearch_dinov2_main/dinov2/configs/eval/vitl14_pretrain.yaml

@@ -0,0 +1,6 @@
+student:
+  arch: vit_large
+  patch_size: 14
+crops:
+  global_crops_size: 518  # this is to set up the position embeddings properly
+  local_crops_size: 98

torchhub/facebookresearch_dinov2_main/dinov2/configs/eval/vits14_pretrain.yaml

@@ -0,0 +1,6 @@
+student:
+  arch: vit_small
+  patch_size: 14
+crops:
+  global_crops_size: 518  # this is to set up the position embeddings properly
+  local_crops_size: 98

torchhub/facebookresearch_dinov2_main/dinov2/configs/ssl_default_config.yaml

@@ -0,0 +1,115 @@
+MODEL:
+  WEIGHTS: ''
+compute_precision:
+  grad_scaler: true
+  teacher:
+    backbone:
+      sharding_strategy: SHARD_GRAD_OP
+      mixed_precision:
+        param_dtype: fp16
+        reduce_dtype: fp16
+        buffer_dtype: fp32
+    dino_head:
+      sharding_strategy: SHARD_GRAD_OP
+      mixed_precision:
+        param_dtype: fp16
+        reduce_dtype: fp16
+        buffer_dtype: fp32
+    ibot_head:
+      sharding_strategy: SHARD_GRAD_OP
+      mixed_precision:
+        param_dtype: fp16
+        reduce_dtype: fp16
+        buffer_dtype: fp32
+  student:
+    backbone:
+      sharding_strategy: SHARD_GRAD_OP
+      mixed_precision:
+        param_dtype: fp16
+        reduce_dtype: fp16
+        buffer_dtype: fp32
+    dino_head:
+      sharding_strategy: SHARD_GRAD_OP
+      mixed_precision:
+        param_dtype: fp16
+        reduce_dtype: fp32
+        buffer_dtype: fp32
+    ibot_head:
+      sharding_strategy: SHARD_GRAD_OP
+      mixed_precision:
+        param_dtype: fp16
+        reduce_dtype: fp32
+        buffer_dtype: fp32
+dino:
+  loss_weight: 1.0
+  head_n_prototypes: 65536
+  head_bottleneck_dim: 256
+  head_nlayers: 3
+  head_hidden_dim: 2048
+  koleo_loss_weight: 0.1
+ibot:
+  loss_weight: 1.0
+  mask_sample_probability: 0.5
+  mask_ratio_min_max:
+  - 0.1
+  - 0.5
+  separate_head: false
+  head_n_prototypes: 65536
+  head_bottleneck_dim: 256
+  head_nlayers: 3
+  head_hidden_dim: 2048
+train:
+  batch_size_per_gpu: 64
+  dataset_path: ImageNet:split=TRAIN
+  output_dir: .
+  saveckp_freq: 20
+  seed: 0
+  num_workers: 10
+  OFFICIAL_EPOCH_LENGTH: 1250
+  cache_dataset: true
+  centering: "centering" # or "sinkhorn_knopp"
+student:
+  arch: vit_large
+  patch_size: 16
+  drop_path_rate: 0.3
+  layerscale: 1.0e-05
+  drop_path_uniform: true
+  pretrained_weights: ''
+  ffn_layer: "mlp"
+  block_chunks: 0
+  qkv_bias: true
+  proj_bias: true
+  ffn_bias: true
+teacher:
+  momentum_teacher: 0.992
+  final_momentum_teacher: 1
+  warmup_teacher_temp: 0.04
+  teacher_temp: 0.07
+  warmup_teacher_temp_epochs: 30
+optim:
+  epochs: 100
+  weight_decay: 0.04
+  weight_decay_end: 0.4
+  base_lr: 0.004  # learning rate for a batch size of 1024
+  lr: 0.  # will be set after applying scaling rule
+  warmup_epochs: 10
+  min_lr: 1.0e-06
+  clip_grad: 3.0
+  freeze_last_layer_epochs: 1
+  scaling_rule: sqrt_wrt_1024
+  patch_embed_lr_mult: 0.2
+  layerwise_decay: 0.9
+  adamw_beta1: 0.9
+  adamw_beta2: 0.999
+crops:
+  global_crops_scale:
+  - 0.32
+  - 1.0
+  local_crops_number: 8
+  local_crops_scale:
+  - 0.05
+  - 0.32
+  global_crops_size: 224
+  local_crops_size: 96
+evaluation:
+  eval_period_iterations: 12500

torchhub/facebookresearch_dinov2_main/dinov2/configs/train/vitg14.yaml

@@ -0,0 +1,26 @@
+dino:
+  head_n_prototypes: 131072
+  head_bottleneck_dim: 384
+ibot:
+  separate_head: true
+  head_n_prototypes: 131072
+train:
+  batch_size_per_gpu: 12
+  dataset_path: ImageNet22k
+  centering: sinkhorn_knopp
+student:
+  arch: vit_giant2
+  patch_size: 14
+  drop_path_rate: 0.4
+  ffn_layer: swiglufused
+  block_chunks: 4
+teacher:
+  momentum_teacher: 0.994
+optim:
+  epochs: 500
+  weight_decay_end: 0.2
+  base_lr: 2.0e-04  # learning rate for a batch size of 1024
+  warmup_epochs: 80
+  layerwise_decay: 1.0
+crops:
+  local_crops_size: 98

torchhub/facebookresearch_dinov2_main/dinov2/configs/train/vitl14.yaml

@@ -0,0 +1,26 @@
+dino:
+  head_n_prototypes: 131072
+  head_bottleneck_dim: 384
+ibot:
+  separate_head: true
+  head_n_prototypes: 131072
+train:
+  batch_size_per_gpu: 32
+  dataset_path: ImageNet22k
+  centering: sinkhorn_knopp
+student:
+  arch: vit_large
+  patch_size: 14
+  drop_path_rate: 0.4
+  ffn_layer: swiglufused
+  block_chunks: 4
+teacher:
+  momentum_teacher: 0.994
+optim:
+  epochs: 500
+  weight_decay_end: 0.2
+  base_lr: 2.0e-04  # learning rate for a batch size of 1024
+  warmup_epochs: 80
+  layerwise_decay: 1.0
+crops:
+  local_crops_size: 98

torchhub/facebookresearch_dinov2_main/dinov2/configs/train/vitl16_short.yaml

@@ -0,0 +1,6 @@
+# this corresponds to the default config
+train:
+  dataset_path: ImageNet:split=TRAIN
+  batch_size_per_gpu: 64
+student:
+  block_chunks: 4

torchhub/facebookresearch_dinov2_main/dinov2/data/__init__.py

@@ -0,0 +1,11 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+from .adapters import DatasetWithEnumeratedTargets
+from .loaders import make_data_loader, make_dataset, SamplerType
+from .collate import collate_data_and_cast
+from .masking import MaskingGenerator
+from .augmentations import DataAugmentationDINO

torchhub/facebookresearch_dinov2_main/dinov2/data/adapters.py

@@ -0,0 +1,29 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+from typing import Any, Tuple
+
+from torch.utils.data import Dataset
+
+
+class DatasetWithEnumeratedTargets(Dataset):
+    def __init__(self, dataset):
+        self._dataset = dataset
+
+    def get_image_data(self, index: int) -> bytes:
+        return self._dataset.get_image_data(index)
+
+    def get_target(self, index: int) -> Tuple[Any, int]:
+        target = self._dataset.get_target(index)
+        return (index, target)
+
+    def __getitem__(self, index: int) -> Tuple[Any, Tuple[Any, int]]:
+        image, target = self._dataset[index]
+        target = index if target is None else target
+        return image, (index, target)
+
+    def __len__(self) -> int:
+        return len(self._dataset)

torchhub/facebookresearch_dinov2_main/dinov2/data/augmentations.py

@@ -0,0 +1,119 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import logging
+
+from torchvision import transforms
+
+from .transforms import (
+    GaussianBlur,
+    make_normalize_transform,
+)
+
+
+logger = logging.getLogger("dinov2")
+
+
+class DataAugmentationDINO(object):
+    def __init__(
+        self,
+        global_crops_scale,
+        local_crops_scale,
+        local_crops_number,
+        global_crops_size=224,
+        local_crops_size=96,
+    ):
+        self.global_crops_scale = global_crops_scale
+        self.local_crops_scale = local_crops_scale
+        self.local_crops_number = local_crops_number
+        self.global_crops_size = global_crops_size
+        self.local_crops_size = local_crops_size
+
+        logger.info("###################################")
+        logger.info("Using data augmentation parameters:")
+        logger.info(f"global_crops_scale: {global_crops_scale}")
+        logger.info(f"local_crops_scale: {local_crops_scale}")
+        logger.info(f"local_crops_number: {local_crops_number}")
+        logger.info(f"global_crops_size: {global_crops_size}")
+        logger.info(f"local_crops_size: {local_crops_size}")
+        logger.info("###################################")
+
+        # random resized crop and flip
+        self.geometric_augmentation_global = transforms.Compose(
+            [
+                transforms.RandomResizedCrop(
+                    global_crops_size, scale=global_crops_scale, interpolation=transforms.InterpolationMode.BICUBIC
+                ),
+                transforms.RandomHorizontalFlip(p=0.5),
+            ]
+        )
+
+        self.geometric_augmentation_local = transforms.Compose(
+            [
+                transforms.RandomResizedCrop(
+                    local_crops_size, scale=local_crops_scale, interpolation=transforms.InterpolationMode.BICUBIC
+                ),
+                transforms.RandomHorizontalFlip(p=0.5),
+            ]
+        )
+
+        # color distorsions / blurring
+        color_jittering = transforms.Compose(
+            [
+                transforms.RandomApply(
+                    [transforms.ColorJitter(brightness=0.4, contrast=0.4, saturation=0.2, hue=0.1)],
+                    p=0.8,
+                ),
+                transforms.RandomGrayscale(p=0.2),
+            ]
+        )
+
+        global_transfo1_extra = GaussianBlur(p=1.0)
+
+        global_transfo2_extra = transforms.Compose(
+            [
+                GaussianBlur(p=0.1),
+                transforms.RandomSolarize(threshold=128, p=0.2),
+            ]
+        )
+
+        local_transfo_extra = GaussianBlur(p=0.5)
+
+        # normalization
+        self.normalize = transforms.Compose(
+            [
+                transforms.ToTensor(),
+                make_normalize_transform(),
+            ]
+        )
+
+        self.global_transfo1 = transforms.Compose([color_jittering, global_transfo1_extra, self.normalize])
+        self.global_transfo2 = transforms.Compose([color_jittering, global_transfo2_extra, self.normalize])
+        self.local_transfo = transforms.Compose([color_jittering, local_transfo_extra, self.normalize])
+
+    def __call__(self, image):
+        output = {}
+
+        # global crops:
+        im1_base = self.geometric_augmentation_global(image)
+        global_crop_1 = self.global_transfo1(im1_base)
+
+        im2_base = self.geometric_augmentation_global(image)
+        global_crop_2 = self.global_transfo2(im2_base)
+
+        output["global_crops"] = [global_crop_1, global_crop_2]
+
+        # global crops for teacher:
+        output["global_crops_teacher"] = [global_crop_1, global_crop_2]
+
+        # local crops:
+        local_crops = [
+            self.local_transfo(self.geometric_augmentation_local(image)) for _ in range(self.local_crops_number)
+        ]
+        output["local_crops"] = local_crops
+        output["offsets"] = ()
+
+        return output

torchhub/facebookresearch_dinov2_main/dinov2/data/collate.py

@@ -0,0 +1,50 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+import random
+
+
+def collate_data_and_cast(samples_list, mask_ratio_tuple, mask_probability, dtype, n_tokens=None, mask_generator=None):
+    # dtype = torch.half  # TODO: Remove
+
+    n_global_crops = len(samples_list[0][0]["global_crops"])
+    n_local_crops = len(samples_list[0][0]["local_crops"])
+
+    collated_global_crops = torch.stack([s[0]["global_crops"][i] for i in range(n_global_crops) for s in samples_list])
+
+    collated_local_crops = torch.stack([s[0]["local_crops"][i] for i in range(n_local_crops) for s in samples_list])
+
+    B = len(collated_global_crops)
+    N = n_tokens
+    n_samples_masked = int(B * mask_probability)
+    probs = torch.linspace(*mask_ratio_tuple, n_samples_masked + 1)
+    upperbound = 0
+    masks_list = []
+    for i in range(0, n_samples_masked):
+        prob_min = probs[i]
+        prob_max = probs[i + 1]
+        masks_list.append(torch.BoolTensor(mask_generator(int(N * random.uniform(prob_min, prob_max)))))
+        upperbound += int(N * prob_max)
+    for i in range(n_samples_masked, B):
+        masks_list.append(torch.BoolTensor(mask_generator(0)))
+
+    random.shuffle(masks_list)
+
+    collated_masks = torch.stack(masks_list).flatten(1)
+    mask_indices_list = collated_masks.flatten().nonzero().flatten()
+
+    masks_weight = (1 / collated_masks.sum(-1).clamp(min=1.0)).unsqueeze(-1).expand_as(collated_masks)[collated_masks]
+
+    return {
+        "collated_global_crops": collated_global_crops.to(dtype),
+        "collated_local_crops": collated_local_crops.to(dtype),
+        "collated_masks": collated_masks,
+        "mask_indices_list": mask_indices_list,
+        "masks_weight": masks_weight,
+        "upperbound": upperbound,
+        "n_masked_patches": torch.full((1,), fill_value=mask_indices_list.shape[0], dtype=torch.long),
+    }

torchhub/facebookresearch_dinov2_main/dinov2/data/datasets/__init__.py

@@ -0,0 +1,8 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+from .image_net import ImageNet
+from .image_net_22k import ImageNet22k

torchhub/facebookresearch_dinov2_main/dinov2/data/datasets/decoders.py

@@ -0,0 +1,32 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+from io import BytesIO
+from typing import Any
+
+from PIL import Image
+
+
+class Decoder:
+    def decode(self) -> Any:
+        raise NotImplementedError
+
+
+class ImageDataDecoder(Decoder):
+    def __init__(self, image_data: bytes) -> None:
+        self._image_data = image_data
+
+    def decode(self) -> Image:
+        f = BytesIO(self._image_data)
+        return Image.open(f).convert(mode="RGB")
+
+
+class TargetDecoder(Decoder):
+    def __init__(self, target: Any):
+        self._target = target
+
+    def decode(self) -> Any:
+        return self._target