Change desc

app.py

@@ -40,10 +40,12 @@ def generate(image, prompt, num_inference_steps=50, image_guidance_scale=1.5, gu
 
 # Update the example list to remove mask-related entries
 example_lists = [
-    ['UltraEdit/images/example_images/1-input.png', "Add a moon in the sky", 20, 1.5, 12.5, 255],
-    ['UltraEdit/images/example_images/2-input.png', "Add cherry blossoms", 20, 1.5, 12.5, 255],
-    ['UltraEdit/images/example_images/3-input.png', "Please dress her in a short purple wedding dress adorned with white floral embroidery.", 20, 1.5, 7.5, 255],
-    ['UltraEdit/images/example_images/4-input.png', "Give her a chief's headdress.", 20, 1.5, 7.5, 24555]
+    ['UltraEdit/images/example_images/4ppl2.jpg', "Add a flower on the t-shirt of the guy in the middle with dark jeans", 50, 1.5, 7.5, 255],
+    ['UltraEdit/images/example_images/cat2.jpg', "Add a green scarf to the right cat", 50, 1.5, 7.5, 255],
+    ['UltraEdit/images/example_images/3ppl2.jpg', "Add a flower bunch to the person with a red jacket", 50, 1.5, 7.5, 255],
+    ['UltraEdit/images/example_images/4ppl1.jpg', "Let the rightmost person wear a golden dress", 50, 1.5, 7.5, 255],
+    ['UltraEdit/images/example_images/bowls1.jpg', "Remove the bowl with some leaves in the middle", 50, 1.5, 7.5, 255],
+    ['UltraEdit/images/example_images/cat1.jpg', "Can we have a dog instead of the cat looking at the camera?", 50, 1.5, 7.5, 255],
 ]
 
 # Update the mask_ex_list to reflect the new example list structure
@@ -67,66 +69,57 @@ outputs = [gr.Image(label="Generated Image")]
 article_html = """
 <div style="text-align: center; max-width: 1000px; margin: 20px auto; font-family: Arial, sans-serif;">
   <h2 style="font-weight: 900; font-size: 2.5rem; margin-bottom: 0.5rem;">
-    🖼️ RefEdit-SD3 for Fine-Grained Image Editing
+    RefEdit-SD3 for Fine-Grained Image Editing
   </h2>
   <div style="margin-bottom: 1rem;">
     <h3 style="font-weight: 500; font-size: 1.25rem; margin: 0;"></h3>
     <p style="font-weight: 400; font-size: 1rem; margin: 0.5rem 0;">
-      Haozhe Zhao<sup>1*</sup>, Xiaojian Ma<sup>2*</sup>, Liang Chen<sup>1</sup>, Shuzheng Si<sup>1</sup>, Rujie Wu<sup>1</sup>,
-      Kaikai An<sup>1</sup>, Peiyu Yu<sup>3</sup>, Minjia Zhang<sup>4</sup>, Qing Li<sup>2</sup>, Baobao Chang<sup>2</sup>
+      Bimsara Pathiraja<sup>1*</sup>, Maitreya Patel<sup>2*</sup>, Shivam Singh, Yezhou Yang, Chitta Baral
     </p>
     <p style="font-weight: 400; font-size: 1rem; margin: 0;">
-      <sup>1</sup>Peking University, <sup>2</sup>BIGAI, <sup>3</sup>UCLA, <sup>4</sup>UIUC
+      Arizona State University
     </p>
   </div>
   <div style="margin: 1rem 0; display: flex; justify-content: center; gap: 1.5rem; flex-wrap: wrap;">
-    <a href="https://huggingface.co/datasets/BleachNick/UltraEdit" style="display: flex; align-items: center; text-decoration: none; color: blue; font-weight: bold; gap: 0.5rem;">
+    <a href="https://huggingface.co/datasets/bpathir1/RefEdit" style="display: flex; align-items: center; text-decoration: none; color: blue; font-weight: bold; gap: 0.5rem;">
       <img src="https://huggingface.co/front/assets/huggingface_logo-noborder.svg" alt="Dataset_4M" style="height: 20px; vertical-align: middle;"> Dataset
     </a>
-    <a href="https://huggingface.co/datasets/BleachNick/UltraEdit_500k" style="display: flex; align-items: center; text-decoration: none; color: blue; font-weight: bold; gap: 0.5rem;">
-      <img src="https://huggingface.co/front/assets/huggingface_logo-noborder.svg" alt="Dataset_500k" style="height: 20px; vertical-align: middle;"> Dataset_500k
-    </a>
-    <a href="https://ultra-editing.github.io/" style="display: flex; align-items: center; text-decoration: none; color: blue; font-weight: bold; gap: 0.5rem;">
+
+    <a href="https://refedit.vercel.app/" style="display: flex; align-items: center; text-decoration: none; color: blue; font-weight: bold; gap: 0.5rem;">
       <span style="font-size: 20px; vertical-align: middle;">🔗</span> Page
     </a>
-    <a href="https://huggingface.co/BleachNick/SD3_UltraEdit_w_mask" style="display: flex; align-items: center; text-decoration: none; color: blue; font-weight: bold; gap: 0.5rem;">
+    <a href="https://huggingface.co/bpathir1/RefEdit-SD3" style="display: flex; align-items: center; text-decoration: none; color: blue; font-weight: bold; gap: 0.5rem;">
       <img src="https://huggingface.co/front/assets/huggingface_logo-noborder.svg" alt="Checkpoint" style="height: 20px; vertical-align: middle;"> Checkpoint
     </a>
-    <a href="https://github.com/HaozheZhao/UltraEdit" style="display: flex; align-items: center; text-decoration: none; color: blue; font-weight: bold; gap: 0.5rem;">
+    <a href="https://github.com/bimsarapathiraja/refedit_private" style="display: flex; align-items: center; text-decoration: none; color: blue; font-weight: bold; gap: 0.5rem;">
       <img src="https://upload.wikimedia.org/wikipedia/commons/9/91/Octicons-mark-github.svg" alt="GitHub" style="height: 20px; vertical-align: middle;"> GitHub
     </a>
   </div>
   <div style="text-align: left; margin: 0 auto; font-size: 1rem; line-height: 1.5;">
     <p>
-      <b>UltraEdit</b> is a dataset designed for fine-grained, instruction-based image editing. It contains over 4 million free-form image editing samples and more than 100,000 region-based image editing samples, automatically generated with real images as anchors.
-    </p>
-    <p>
-      This demo allows you to perform image editing using the <a href="https://huggingface.co/stabilityai/stable-diffusion-3-medium-diffusers" style="color: blue; text-decoration: none;">Stable Diffusion 3</a> model trained with this extensive dataset. It supports both free-form (without mask) and region-based (with mask) image editing. Use the sliders to adjust the inference steps and guidance scales, and provide a seed for reproducibility. The image guidance scale of 1.5 and text guidance scale of 7.5 / 12.5 is a good start for free-form/region-based image editing.
+      <b>RefEdit</b> is a benchmark and method for improving instruction-based image editing models for referring expressions.
     </p>
-    <p>
-     <b>Usage Instructions:</b> You need to upload the images and prompts for editing. Use the pen tool to mark the areas you want to edit. If no region is marked, it will resort to free-form editing.
-     </p>
   </div>
 </div>
 """
-html = '''
-  <div style="text-align: left; margin-top: 2rem; font-size: 0.85rem; color: gray;">
-    <b>Limitations:</b>
-    <ul>
-      <li>We have not conducted any NSFW checks;</li>
-      <li>Due to the bias of the generated models, the model performance is still weak when dealing with high-frequency information such as <b>human facial expressions or text in the images</b>;</li>
-      <li>We unified the free-form and region-based image editing by adding an extra channel of the mask image to the dataset. When doing free-form image editing, the network receives a blank mask.</li>
-      <li>The generation result is sensitive to the guidance scale. For text guidance, based on experience, free-form image editing will perform better with a relatively low guidance score (7.5 or lower), while region-based image editing will perform better with a higher guidance score.</li>
-    </ul>
-  </div>
-'''
+# html = '''
+#   <div style="text-align: left; margin-top: 2rem; font-size: 0.85rem; color: gray;">
+#     <b>Limitations:</b>
+#     <ul>
+#       <li>We have not conducted any NSFW checks;</li>
+#       <li>Due to the bias of the generated models, the model performance is still weak when dealing with high-frequency information such as <b>human facial expressions or text in the images</b>;</li>
+#       <li>We unified the free-form and region-based image editing by adding an extra channel of the mask image to the dataset. When doing free-form image editing, the network receives a blank mask.</li>
+#       <li>The generation result is sensitive to the guidance scale. For text guidance, based on experience, free-form image editing will perform better with a relatively low guidance score (7.5 or lower), while region-based image editing will perform better with a higher guidance score.</li>
+#     </ul>
+#   </div>
+# '''
 
 demo = gr.Interface(
     fn=generate,
     inputs=inputs,
     outputs=outputs,
     description=article_html,
-    article=html,
+    # article=html,
     examples=mask_ex_list,
     cache_examples = True,
     live = False