add files.

app.py

@@ -1,103 +1,262 @@
-from constants import css 
-
-import gradio as gr 
-
-
-block = gr.Blocks(css=css)
-
-with block:
-    gr.HTML(
-        """
-            <div style="text-align: center; margin: 0 auto;">
-              <div
-                style="
-                  display: inline-flex;
-                  align-items: center;
-                  gap: 0.8rem;
-                  font-size: 1.75rem;
-                "
-              >
-                <svg
-                  width="0.65em"
-                  height="0.65em"
-                  viewBox="0 0 115 115"
-                  fill="none"
-                  xmlns="http://www.w3.org/2000/svg"
-                >
-                  <rect width="23" height="23" fill="white"></rect>
-                  <rect y="69" width="23" height="23" fill="white"></rect>
-                  <rect x="23" width="23" height="23" fill="#AEAEAE"></rect>
-                  <rect x="23" y="69" width="23" height="23" fill="#AEAEAE"></rect>
-                  <rect x="46" width="23" height="23" fill="white"></rect>
-                  <rect x="46" y="69" width="23" height="23" fill="white"></rect>
-                  <rect x="69" width="23" height="23" fill="black"></rect>
-                  <rect x="69" y="69" width="23" height="23" fill="black"></rect>
-                  <rect x="92" width="23" height="23" fill="#D9D9D9"></rect>
-                  <rect x="92" y="69" width="23" height="23" fill="#AEAEAE"></rect>
-                  <rect x="115" y="46" width="23" height="23" fill="white"></rect>
-                  <rect x="115" y="115" width="23" height="23" fill="white"></rect>
-                  <rect x="115" y="69" width="23" height="23" fill="#D9D9D9"></rect>
-                  <rect x="92" y="46" width="23" height="23" fill="#AEAEAE"></rect>
-                  <rect x="92" y="115" width="23" height="23" fill="#AEAEAE"></rect>
-                  <rect x="92" y="69" width="23" height="23" fill="white"></rect>
-                  <rect x="69" y="46" width="23" height="23" fill="white"></rect>
-                  <rect x="69" y="115" width="23" height="23" fill="white"></rect>
-                  <rect x="69" y="69" width="23" height="23" fill="#D9D9D9"></rect>
-                  <rect x="46" y="46" width="23" height="23" fill="black"></rect>
-                  <rect x="46" y="115" width="23" height="23" fill="black"></rect>
-                  <rect x="46" y="69" width="23" height="23" fill="black"></rect>
-                  <rect x="23" y="46" width="23" height="23" fill="#D9D9D9"></rect>
-                  <rect x="23" y="115" width="23" height="23" fill="#AEAEAE"></rect>
-                  <rect x="23" y="69" width="23" height="23" fill="black"></rect>
-                </svg>
-                <h1 style="font-weight: 900; margin-bottom: 7px;margin-top:5px">
-                  Stable Diffusion 2.1 Demo
-                </h1>
-              </div>
-              <p style="margin-bottom: 10px; font-size: 94%; line-height: 23px;">
-                Stable Diffusion 2.1 is the latest text-to-image model from StabilityAI. <a style="text-decoration: underline;" href="https://huggingface.co/spaces/stabilityai/stable-diffusion-1">Access Stable Diffusion 1 Space here</a><br>For faster generation and API
-                access you can try
-                <a
-                  href="http://beta.dreamstudio.ai/"
-                  style="text-decoration: underline;"
-                  target="_blank"
-                  >DreamStudio Beta</a
-                >.</a>
-              </p>
-            </div>
-        """
+import importlib
+from functools import partial
+from typing import List
+
+import gradio as gr
+import numpy as np
+import torch
+from diffusers import StableDiffusionPipeline
+from PIL import Image
+from torchmetrics.functional.multimodal import clip_score
+from torchmetrics.image.inception import InceptionScore
+
+SEED = 0
+WEIGHT_DTYPE = torch.float16
+
+TITLE = "Evaluate Schedulers with StableDiffusionPipeline 🧨"
+DESCRIPTION = """
+This Space allows you to quantitatively compare [different noise schedulers](https://huggingface.co/docs/diffusers/using-diffusers/schedulers) with a [`StableDiffusionPipeline`](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/overview).
+
+One of the applications of this Space could be to evaluate different schedulers for a certain Stable Diffusion checkpoint for a fixed number of inference steps.
+
+Here's how it works:
+
+* The evaluator first sets a seed and then generates the initial noise which is passed as the initial latent to start the image generation process. It is done to ensure fair comparison.
+* This initial latent is used every time the pipeline is run (with different schedulers).
+* To quantify the quality of the generated images we use:
+    * [Inception Score](https://en.wikipedia.org/wiki/Inception_score)
+    * [Clip Score](https://arxiv.org/abs/2104.08718)
+
+**Notes**:
+
+* The default scheduler associated with the provided checkpoint is always used for reporting the scores.
+* Increasing both the number of images per prompt and the number of inference steps could quickly build up the inference queue and thus
+resulting in slowdowns.
+"""
+
+
+inception_score_fn = InceptionScore(normalize=True)
+torch.manual_seed(SEED)
+clip_score_fn = partial(clip_score, model_name_or_path="openai/clip-vit-base-patch16")
+
+
+def make_grid(images, rows, cols):
+    w, h = images[0].size
+    grid = Image.new("RGB", size=(cols * w, rows * h))
+    for i, image in enumerate(images):
+        grid.paste(image, box=(i % cols * w, i // cols * h))
+    return grid
+
+
+# Copied from https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_utils.py#L814
+def numpy_to_pil(images):
+    """
+    Convert a numpy image or a batch of images to a PIL image.
+    """
+    if images.ndim == 3:
+        images = images[None, ...]
+    images = (images * 255).round().astype("uint8")
+    if images.shape[-1] == 1:
+        # special case for grayscale (single channel) images
+        pil_images = [Image.fromarray(image.squeeze(), mode="L") for image in images]
+    else:
+        pil_images = [Image.fromarray(image) for image in images]
+
+    return pil_images
+
+
+def prepare_report(scheduler_name: str, results: dict):
+    image_grid = results["images"]
+    scores = results["scores"]
+    img_str = ""
+
+    image_name = f"{scheduler_name}_images.png"
+    image_grid.save(image_name)
+    img_str = img_str = f"![img_grid_{scheduler_name}](/file=./{image_name})\n"
+
+    report_str = f"""
+\n\n## {scheduler_name}
+
+### Sample images
+
+{img_str}
+
+### Scores
+
+{scores}
+\n\n
+    """
+
+    return report_str
+
+
+def initialize_pipeline(checkpoint: str):
+    sd_pipe = StableDiffusionPipeline.from_pretrained(
+        checkpoint, torch_dtype=WEIGHT_DTYPE
+    )
+    sd_pipe = sd_pipe.to("cuda")
+    original_scheduler_config = sd_pipe.scheduler.config
+    return sd_pipe, original_scheduler_config
+
+
+def get_scheduler(scheduler_name: str):
+    schedulers_lib = importlib.import_module("diffusers", package="schedulers")
+    scheduler_abs = getattr(schedulers_lib, scheduler_name)
+
+    return scheduler_abs
+
+
+def get_latents(num_images_per_prompt: int, seed=SEED):
+    generator = torch.manual_seed(seed)
+    latents = np.random.RandomState(seed).standard_normal(
+        (num_images_per_prompt, 4, 64, 64)
     )
-    with gr.Group():
-        with gr.Box():
-            with gr.Row(elem_id="prompt-container").style(mobile_collapse=False, equal_height=True):
-                with gr.Column():
-                    text = gr.Textbox(
-                        label="Enter your prompt",
-                        show_label=False,
-                        max_lines=1,
-                        placeholder="Enter your prompt",
-                        elem_id="prompt-text-input",
-                    ).style(
-                        border=(True, False, True, True),
-                        rounded=(True, False, False, True),
-                        container=False,
-                    )
-                    negative = gr.Textbox(
-                        label="Enter your negative prompt",
-                        show_label=False,
-                        max_lines=1,
-                        placeholder="Enter a negative prompt",
-                        elem_id="negative-prompt-text-input",
-                    ).style(
-                        border=(True, False, True, True),
-                        rounded=(True, False, False, True),
-                        container=False,
-                    )
-                btn = gr.Button("Generate image").style(
-                    margin=False,
-                    rounded=(False, True, True, False),
-                    full_width=False,
-                )
-
-
-block.launch()
+    latents = torch.from_numpy(latents).to(device="cuda", dtype=WEIGHT_DTYPE)
+    return latents
+
+
+def compute_metrics(images: np.ndarray, prompts: List[str]):
+    inception_score_fn.update(torch.from_numpy(images).permute(0, 3, 1, 2))
+    inception_score = inception_score_fn.compute()
+
+    images_int = (images * 255).astype("uint8")
+    clip_score = clip_score_fn(
+        torch.from_numpy(images_int).permute(0, 3, 1, 2), prompts
+    ).detach()
+    return {
+        "inception_score (⬆️)": {
+            "mean": round(float(inception_score[0]), 4),
+            "std": round(float(inception_score[1]), 4),
+        },
+        "clip_score (⬆️)": round(float(clip_score), 4),
+    }
+
+
+def run(
+    prompt: str,
+    num_images_per_prompt: int,
+    num_inference_steps: int,
+    checkpoint: str,
+    schedulers_to_test: List[str],
+):
+    all_images = {}
+
+    sd_pipeline, original_scheduler_config = initialize_pipeline(checkpoint)
+    latents = get_latents(num_images_per_prompt)
+    prompts = [prompt] * num_images_per_prompt
+
+    images = sd_pipeline(
+        prompts,
+        latents=latents,
+        num_inference_steps=num_inference_steps,
+        output_type="numpy",
+    ).images
+    original_scheduler_name = original_scheduler_config._class_name
+
+    all_images.update(
+        {
+            original_scheduler_name: {
+                "images": make_grid(numpy_to_pil(images), 1, num_images_per_prompt),
+                "scores": compute_metrics(images, prompts),
+            }
+        }
+    )
+    # print("First scheduler complete.")
+
+    for scheduler_name in schedulers_to_test:
+        if scheduler_name == original_scheduler_name:
+            continue
+        scheduler_cls = get_scheduler(scheduler_name)
+        current_scheduler = scheduler_cls.from_config(original_scheduler_config)
+        sd_pipeline.scheduler = current_scheduler
+
+        cur_scheduler_images = sd_pipeline(
+            prompts, num_inference_steps=num_inference_steps, output_type="numpy"
+        ).images
+        all_images.update(
+            {
+                scheduler_name: {
+                    "images": make_grid(
+                        numpy_to_pil(cur_scheduler_images), 1, num_images_per_prompt
+                    ),
+                    "scores": compute_metrics(cur_scheduler_images, prompts),
+                }
+            }
+        )
+        # print(f"{scheduler_name} complete.")
+
+    output_str = ""
+    for scheduler_name in all_images:
+        # print(f"scheduler_name: {scheduler_name}")
+        output_str += prepare_report(scheduler_name, all_images[scheduler_name])
+    # print(output_str)
+    return output_str
+
+
+demo = gr.Interface(
+    run,
+    inputs=[
+        gr.Text(max_lines=1, placeholder="a painting of a dog"),
+        gr.Slider(3, 10, value=3, step=1),
+        gr.Slider(10, 100, value=50, step=1),
+        gr.Dropdown(
+            [
+                "CompVis/stable-diffusion-v1-4",
+                "runwayml/stable-diffusion-v1-5",
+                "stabilityai/stable-diffusion-2-base",
+            ],
+            value="CompVis/stable-diffusion-v1-4",
+            multiselect=False,
+            interactive=True,
+        ),
+        gr.Dropdown(
+            [
+                "EulerDiscreteScheduler",
+                "PNDMScheduler",
+                "LMSDiscreteScheduler",
+                "DPMSolverMultistepScheduler",
+                "DDIMScheduler",
+            ],
+            value=["LMSDiscreteScheduler"],
+            multiselect=True,
+        ),
+    ],
+    outputs=[gr.Markdown().style()],
+    title=TITLE,
+    description=DESCRIPTION,
+    allow_flagging=False,
+)
+demo.launch()
+
+with gr.Blocks() as demo:
+    with gr.Row():
+        with gr.Column():
+            prompt = gr.Text(max_lines=1, placeholder="a painting of a dog")
+            num_images_per_prompt = gr.Slider(3, 10, value=3, step=1)
+            num_inference_steps = gr.Slider(10, 100, value=50, step=1)
+            model_ckpt = gr.Dropdown(
+                [
+                    "CompVis/stable-diffusion-v1-4",
+                    "runwayml/stable-diffusion-v1-5",
+                    "stabilityai/stable-diffusion-2-base",
+                    "Other"
+                ],
+                value="CompVis/stable-diffusion-v1-4",
+                multiselect=False,
+                interactive=True,
+            )
+            other_finedtuned_checkpoints = gr.Textbox(visible=False)
+            model_ckpt.change(lambda x: gr.Dropdown.update(visible=x=="Other"), model_ckpt, other_finedtuned_checkpoints)
+            schedulers_to_test = gr.Dropdown(
+                [
+                    "EulerDiscreteScheduler",
+                    "PNDMScheduler",
+                    "LMSDiscreteScheduler",
+                    "DPMSolverMultistepScheduler",
+                    "DDIMScheduler",
+                ],
+                value=["LMSDiscreteScheduler"],
+                multiselect=True,
+            )
+           
+demo.launch()

constants.py

@@ -1,123 +0,0 @@
-css = """
-        .gradio-container {
-            font-family: 'IBM Plex Sans', sans-serif;
-        }
-        .gr-button {
-            color: white;
-            border-color: black;
-            background: black;
-        }
-        input[type='range'] {
-            accent-color: black;
-        }
-        .dark input[type='range'] {
-            accent-color: #dfdfdf;
-        }
-        .container {
-            max-width: 730px;
-            margin: auto;
-            padding-top: 1.5rem;
-        }
-        #gallery {
-            min-height: 22rem;
-            margin-bottom: 15px;
-            margin-left: auto;
-            margin-right: auto;
-            border-bottom-right-radius: .5rem !important;
-            border-bottom-left-radius: .5rem !important;
-        }
-        #gallery>div>.h-full {
-            min-height: 20rem;
-        }
-        .details:hover {
-            text-decoration: underline;
-        }
-        .gr-button {
-            white-space: nowrap;
-        }
-        .gr-button:focus {
-            border-color: rgb(147 197 253 / var(--tw-border-opacity));
-            outline: none;
-            box-shadow: var(--tw-ring-offset-shadow), var(--tw-ring-shadow), var(--tw-shadow, 0 0 #0000);
-            --tw-border-opacity: 1;
-            --tw-ring-offset-shadow: var(--tw-ring-inset) 0 0 0 var(--tw-ring-offset-width) var(--tw-ring-offset-color);
-            --tw-ring-shadow: var(--tw-ring-inset) 0 0 0 calc(3px var(--tw-ring-offset-width)) var(--tw-ring-color);
-            --tw-ring-color: rgb(191 219 254 / var(--tw-ring-opacity));
-            --tw-ring-opacity: .5;
-        }
-        #advanced-btn {
-            font-size: .7rem !important;
-            line-height: 19px;
-            margin-top: 12px;
-            margin-bottom: 12px;
-            padding: 2px 8px;
-            border-radius: 14px !important;
-        }
-        #advanced-options {
-            display: none;
-            margin-bottom: 20px;
-        }
-        .footer {
-            margin-bottom: 45px;
-            margin-top: 35px;
-            text-align: center;
-            border-bottom: 1px solid #e5e5e5;
-        }
-        .footer>p {
-            font-size: .8rem;
-            display: inline-block;
-            padding: 0 10px;
-            transform: translateY(10px);
-            background: white;
-        }
-        .dark .footer {
-            border-color: #303030;
-        }
-        .dark .footer>p {
-            background: #0b0f19;
-        }
-        .acknowledgments h4{
-            margin: 1.25em 0 .25em 0;
-            font-weight: bold;
-            font-size: 115%;
-        }
-        .animate-spin {
-            animation: spin 1s linear infinite;
-        }
-        @keyframes spin {
-            from {
-                transform: rotate(0deg);
-            }
-            to {
-                transform: rotate(360deg);
-            }
-        }
-        #share-btn-container {
-            display: flex; padding-left: 0.5rem !important; padding-right: 0.5rem !important; background-color: #000000; justify-content: center; align-items: center; border-radius: 9999px !important; width: 13rem;
-            margin-top: 10px;
-            margin-left: auto;
-        }
-        #share-btn {
-            all: initial; color: #ffffff;font-weight: 600; cursor:pointer; font-family: 'IBM Plex Sans', sans-serif; margin-left: 0.5rem !important; padding-top: 0.25rem !important; padding-bottom: 0.25rem !important;right:0;
-        }
-        #share-btn * {
-            all: unset;
-        }
-        #share-btn-container div:nth-child(-n+2){
-            width: auto !important;
-            min-height: 0px !important;
-        }
-        #share-btn-container .wrap {
-            display: none !important;
-        }
-        
-        .gr-form{
-            flex: 1 1 50%; border-top-right-radius: 0; border-bottom-right-radius: 0;
-        }
-        #prompt-container{
-            gap: 0;
-        }
-        #prompt-text-input, #negative-prompt-text-input{padding: .45rem 0.625rem}
-        #component-16{border-top-width: 1px!important;margin-top: 1em}
-        .image_duplication{position: absolute; width: 100px; left: 50px}
-"""

requirements.txt

@@ -0,0 +1,5 @@
+torchmetrics[image]
+transformers 
+diffusers
+accelerate
+numpy