<feat> optimize output format and add examples

.gitignore

@@ -1 +1,3 @@
-__pycache__/
+__pycache__/
+*.jpg
+*.png

app.py

@@ -9,6 +9,8 @@ import copy
 import cv2
 import spaces
 import gc
+import tempfile
+import imageio
 import gradio as gr
 import numpy as np
 
@@ -285,8 +287,167 @@ def process_image_and_text(condition_image, target_prompt, condition_image_promp
         out = (out * 255).astype(np.uint8)
         out = Image.fromarray(out)
         output_images.append(out)
-    
-    return output_images[1:] + [output_images[0]]
+
+    # video = [np.array(img.convert('RGB')) for img in output_images[1:] + [output_images[0]]]
+    # video = np.stack(video, axis=0)
+
+    with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as f:
+        video_path = f.name
+    imageio.mimsave(video_path, output_images[1:]+[output_images[0]], fps=5)
+
+    return output_images[0], video_path
+
+def get_samples():
+    sample_list = [
+        {
+            "task": "subject_driven", 
+            "input": "assets/subject_driven_image_generation_dreambench_input.jpg", 
+            "target_prompt": "a cat in a chef outfit", 
+            "condition_image_prompt": "a cat", 
+            "output": "assets/subject_driven_image_generation_dreambench_output.png", 
+            "inpainting": False,
+            "fill_x1": None,
+            "fill_x2": None,
+            "fill_y1": None,
+            "fill_y2": None,
+        },
+        {
+            "task": "subject_driven", 
+            "input": "assets/subject_driven_image_generation_input.jpg", 
+            "target_prompt": "The woman stands in a snowy forest, captured in a half-portrait", 
+            "condition_image_prompt": "Woman in cream knit sweater sits calmly by a crackling fireplace, surrounded by warm candlelight and rustic wooden shelves", 
+            "output": "assets/subject_driven_image_generation_output.png", 
+            "inpainting": False,
+            "fill_x1": None,
+            "fill_x2": None,
+            "fill_y1": None,
+            "fill_y2": None,
+        },
+        {
+            "task": "canny", 
+            "input": "assets/canny_to_image_input.jpg", 
+            "target_prompt": "Mosquito frozen in clear ice cube on sand, glowing sunset casting golden light with misty halo around ice", 
+            "condition_image_prompt": "", 
+            "output": "assets/canny_to_image_output.png", 
+            "inpainting": False,
+            "fill_x1": None,
+            "fill_x2": None,
+            "fill_y1": None,
+            "fill_y2": None,
+        },
+        {
+            "task": "coloring", 
+            "input": "assets/colorization_input.jpg", 
+            "target_prompt": "A vibrant young woman with rainbow glasses, yellow eyes, and colorful feather accessory against a bright yellow background", 
+            "condition_image_prompt": "", 
+            "output": "assets/colorization_output.png", 
+            "inpainting": False,
+            "fill_x1": None,
+            "fill_x2": None,
+            "fill_y1": None,
+            "fill_y2": None,
+        },
+        {
+            "task": "deblurring", 
+            "input": "assets/deblurring_input.jpg", 
+            "target_prompt": "Vibrant rainbow ball creates dramatic splash in clear water, bubbles swirling against crisp white background", 
+            "condition_image_prompt": "", 
+            "output": "assets/deblurring_output.png", 
+            "inpainting": False,
+            "fill_x1": None,
+            "fill_x2": None,
+            "fill_y1": None,
+            "fill_y2": None,
+        },
+        {
+            "task": "depth", 
+            "input": "assets/depth_to_image_input.jpg", 
+            "target_prompt": "Golden-brown cat-shaped bread loaf with closed eyes rests on wooden table, soft kitchen blur in background", 
+            "condition_image_prompt": "", 
+            "output": "assets/depth_to_image_output.png", 
+            "inpainting": False,
+            "fill_x1": None,
+            "fill_x2": None,
+            "fill_y1": None,
+            "fill_y2": None,
+        },
+        {
+            "task": "depth_pred", 
+            "input": "assets/depth_prediction_input.jpg", 
+            "target_prompt": "Steaming bowl of ramen with pork slices, soft-boiled egg, greens, and scallions in rich broth on wooden table", 
+            "condition_image_prompt": "", 
+            "output": "assets/depth_prediction_output.png", 
+            "inpainting": False,
+            "fill_x1": None,
+            "fill_x2": None,
+            "fill_y1": None,
+            "fill_y2": None,
+        },
+        {
+            "task": "fill", 
+            "input": "assets/inpainting_input.jpg", 
+            "target_prompt": "Mona Lisa dons a medical mask, her enigmatic smile now concealed beneath crisp white fabric", 
+            "condition_image_prompt": "", 
+            "output": "assets/inpainting_output.png", 
+            "inpainting": True,
+            "fill_x1": 170,
+            "fill_x2": 300,
+            "fill_y1": 190,
+            "fill_y2": 290,
+        },
+        {
+            "task": "fill", 
+            "input": "assets/outpainting_input.jpg", 
+            "target_prompt": "Her left hand emerges at the frame's lower right, delicately cradling a vibrant red flower against the black void", 
+            "condition_image_prompt": "", 
+            "output": "assets/outpainting_output.png", 
+            "inpainting": False,
+            "fill_x1": 155,
+            "fill_x2": 512,
+            "fill_y1": 0,
+            "fill_y2": 330,
+        },
+        {
+            "task": "sr", 
+            "input": "assets/super_resolution_input.jpg", 
+            "target_prompt": "Crispy buffalo wings and golden fries rest on a red-and-white checkered paper lining a gleaming metal tray, with creamy dip", 
+            "condition_image_prompt": "", 
+            "output": "assets/super_resolution_output.png", 
+            "inpainting": False,
+            "fill_x1": None,
+            "fill_x2": None,
+            "fill_y1": None,
+            "fill_y2": None,
+        },
+        {
+            "task": "style_transfer", 
+            "input": "assets/style_transfer_input.png", 
+            "target_prompt": "bitmoji style. An orange cat sits quietly on the stone slab. Beside it are the green grasses. With its ears perked up, it looks to one side.", 
+            "condition_image_prompt": "An orange cat sits quietly on the stone slab. Beside it are the green grasses. With its ears perked up, it looks to one side.", 
+            "output": "assets/style_transfer_output.png", 
+            "inpainting": False,
+            "fill_x1": None,
+            "fill_x2": None,
+            "fill_y1": None,
+            "fill_y2": None,
+        },
+    ]
+
+    return [
+        [
+            sample['task'],
+            Image.open(sample['input']),
+            sample['target_prompt'],
+            sample['condition_image_prompt'],
+            Image.open(sample['output']),
+            sample['inpainting'],
+            sample['fill_x1'],
+            sample['fill_x2'],
+            sample['fill_y1'],
+            sample['fill_y2'],
+        ]
+        for sample in sample_list
+    ]
 
 def create_app():
     with gr.Blocks() as app:
@@ -315,7 +476,8 @@ def create_app():
                 )
                 gr.Markdown(notice, elem_id="notice")
                 target_prompt = gr.Textbox(lines=2, label="Target Prompt", elem_id="tp")
-                condition_image_prompt = gr.Textbox(lines=2, label="Condition Image Prompt (Only required by Subject-driven Image Generation and Style Transfer tasks)", elem_id="cp")
+                gr.Markdown("**Condition Image Prompt** _(Only required by Subject-driven Image Generation and Style Transfer tasks)_")
+                condition_image_prompt = gr.Textbox(lines=2, label="Condition Image Prompt", elem_id="cp")
                 random_seed = gr.Number(label="Random Seed", precision=0, value=0, elem_id="seed")
                 num_steps = gr.Number(label="Diffusion Inference Steps", precision=0, value=50, elem_id="steps")
                 inpainting = gr.Checkbox(label="Inpainting", value=False, elem_id="inpainting")
@@ -327,20 +489,33 @@ def create_app():
 
             with gr.Column(variant="panel", elem_classes="outputPanel"):
                 # output_image = gr.Image(type="pil", elem_id="output")
-                output_images = gr.Gallery(
-                    label="Output Images",
-                    show_label=True,
-                    elem_id="output_gallery",
-                    columns=1,
-                    rows=10,
-                    object_fit="contain",
-                    height="auto",
+                # output_images = gr.Gallery(
+                #     label="Output Images",
+                #     show_label=True,
+                #     elem_id="output_gallery",
+                #     columns=1,
+                #     rows=10,
+                #     object_fit="contain",
+                #     height="auto",
+                # )
+                output_image = gr.Image(
+                    type="pil", label="Output Image", elem_id="output_image"
+                )
+                output_video = gr.Video(
+                    label="Output Video", elem_id="output_video"
                 )
 
+        with gr.Row():
+            examples = gr.Examples(
+                examples=get_samples(),
+                inputs=[task, condition_image, target_prompt, condition_image_prompt, output_image, inpainting, fill_x1, fill_x2, fill_y1, fill_y2],
+                label="Examples",
+            )
+
         submit_btn.click(
             fn=process_image_and_text,
             inputs=[condition_image, target_prompt, condition_image_prompt, task, random_seed, num_steps, inpainting, fill_x1, fill_x2, fill_y1, fill_y2],
-            outputs=output_images,
+            outputs=[output_image, output_video],
         )
         
     return app

assets/README.md

@@ -0,0 +1 @@
+Here are some examples.