Update app.py

app.py

@@ -10,7 +10,9 @@ import os
 from PIL import Image, ImageDraw, ImageFont
 import numpy as np
 import hashlib
-import spaces
+import spaces # Make sure to import spaces
+
+
 
 #First output the thinking process in <think> </think> tags and then output the final answer in <answer> </answer> tags.
 #Answer the following question based on the information above and the given image, and provide citations for your response.
@@ -133,8 +135,8 @@ def fetch_wikipedia_info(url):
 def recognize_celebrities(image_path, confidence_threshold=90):
     client = boto3.client(
         "rekognition",
-        aws_access_key_id='AKIA6GSNGQMOW2UX3E52',
-        aws_secret_access_key='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',
+        aws_access_key_id=os.getenv('aws_access_key_id'),
+        aws_secret_access_key=os.getenv('aws_secret_access_key'),
         region_name='us-east-1'
     )
     
@@ -169,7 +171,7 @@ def draw_bounding_boxes(image_path, bounding_boxes, names):
         
         # Add name label
         text = f"[{i+1}]: {names[i]}"
-        font = ImageFont.truetype("arial.ttf", 20)  # Adjust font and size as needed
+        font = ImageFont.truetype("./Arial.ttf", 20)  # Adjust font and size as needed
         text_bbox = draw.textbbox((left, top - 20), text, font=font)
         draw.rectangle(text_bbox, fill="white")
         draw.text((left, top - 20), text, fill="red", font=font)
@@ -190,7 +192,7 @@ model = Qwen2VLForConditionalGeneration.from_pretrained(
     torch_dtype=torch.float16, 
     device_map="auto"
 )
-processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-2B-Instruct")
+processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-7B-Instruct")
 
 # Use Qwen model for Q&A
 def qwen_qa(image_path, question, names, bounding_boxes, en_wiki_pedia_links, en_intros):
@@ -261,6 +263,7 @@ def is_example_image(image, examples):
     return False, None
 
 # Main processing function
+# This is the function that will run on the GPU
 @spaces.GPU
 def process_image(image, question, confidence_threshold, examples=None):
     if image is None:
@@ -335,7 +338,6 @@ def process_image(image, question, confidence_threshold, examples=None):
                 print(links)
                 en_wiki_pedia_links.append("No English Wikipedia link found")
                 en_intros.append(f"This person is {name}.")
-        
         if not names:
             if os.path.exists(temp_image_path):
                 os.remove(temp_image_path)
@@ -403,19 +405,6 @@ def create_interface():
           {'Width': 0.07422236353158951, 'Height': 0.15943190455436707, 'Left': 0.4633428454399109, 'Top': 0.07901764661073685}, 
           {'Width': 0.07562466710805893, 'Height': 0.13936467468738556, 'Left': 0.025178398936986923, 'Top': 0.4953641891479492}], 
          ['https://www.wikidata.org/wiki/Q3572699', 'https://www.wikidata.org/wiki/Q92894', 'https://www.wikidata.org/wiki/Q3571662']],
-        """["./clinton.jpg", "Who are the people in the picture, and what is the relationship between them?", 
-         ['Bill Clinton', 'Monica Lewinsky'], 
-         [{'Width': 0.07620880007743835, 'Height': 0.16198107600212097, 'Left': 0.5074607729911804, 'Top': 0.14220821857452393}, 
-          {'Width': 0.0722670778632164, 'Height': 0.1512720286846161, 'Left': 0.3914872407913208, 'Top': 0.24376636743545532}], 
-         ['https://www.wikidata.org/wiki/Q1124', 'https://www.wikidata.org/wiki/Q212659']],
-        ["./epst.jpeg", "Provide image description.", 
-         ['Lisa Randall', 'Kip S. Thorne', 'David Gross', 'Stephen Hawking', 'Brenda Chapman'], 
-         [{'Width': 0.09916354715824127, 'Height': 0.166521355509758, 'Left': 0.7962431311607361, 'Top': 0.4121580123901367}, 
-          {'Width': 0.07940348237752914, 'Height': 0.1626593917608261, 'Left': 0.6891748905181885, 'Top': 0.33117005228996277}, 
-          {'Width': 0.06350294500589371, 'Height': 0.12757645547389984, 'Left': 0.544218122959137, 'Top': 0.1575603038072586}, 
-          {'Width': 0.06830617040395737, 'Height': 0.1128319799900055, 'Left': 0.2937725782394409, 'Top': 0.30404558777809143}, 
-          {'Width': 0.03966952860355377, 'Height': 0.11658532917499542, 'Left': 0.18093101680278778, 'Top': 0.31299835443496704}], 
-         ['https://www.wikidata.org/wiki/Q450404', 'https://www.wikidata.org/wiki/Q323320', 'https://www.wikidata.org/wiki/Q40262', 'https://www.wikidata.org/wiki/Q17714', 'https://www.wikidata.org/wiki/Q429715']]"""
     ]
     
     # Filter examples to only include files that exist
@@ -424,19 +413,19 @@ def create_interface():
         if os.path.exists(example[0]):
             existing_examples.append(example)
     
-    with gr.Blocks(title="Celebrity Recognition and Q&A System") as app:
-        gr.Markdown("<div style='text-align: center;'><h1 style=' font-size: 28px; '>Celebrity Recognition and Q&A System</h1></div>")
+    with gr.Blocks(title="Celebrity Recognition and VQA System") as app:
+        gr.Markdown("<div style='text-align: center;'><h1 style=' font-size: 28px; '>Celebrity Recognition and VQA</h1></div>")
         gr.Markdown("**RC-MLLM** model is developed based on the Qwen2-VL model through a novel method called **RCVIT (Region-level Context-aware Visual Instruction Tuning)**, using the specially constructed **RCMU dataset** for training. Its core feature is the capability for **Region-level Context-aware Multimodal Understanding (RCMU)**. This means it can simultaneously understand both the visual content of specific regions/objects within an image and their associated textual information (utilizing bounding boxes coordinates), allowing it to respond to user instructions in a more context-aware manner. Simply put, RC-MLLM not only understands images but can also integrate the textual information linked to specific objects within the image for understanding. It achieves outstanding performance on RCMU tasks and is suitable for applications like personalized conversation.")
 
         markdown_content = """
-            📑 [Arxiv](https://arxiv.org/abs/your-paper-id) | 
-            🤗 [Checkpoint]() | 
+            📑 [Region-Level Context-Aware Multimodal Understanding](https://arxiv.org/abs/2508.12263) | 
+            🤗 Models:[RC-Qwen2VL-2b](https://huggingface.co/weihongliang/RC-Qwen2VL-2b/blob/main/README.md) [RC-Qwen2VL-7b](https://huggingface.co/weihongliang/RC-Qwen2VL-7b/blob/main/README.md)| 
             📁 [Dataset](https://huggingface.co/your-model-name) |
-            [Github](https://github.com/your-username/your-repo) | 
-            🚀 [Personalized Conversation](https://your-project-url.com)
+            [Github](https://github.com/hongliang-wei/RC-MLLM) | 
+            🚀 [Personalized Conversation Demo](https://1684c5f6e1c5a19b2c.gradio.live/)
             """
         gr.Markdown(markdown_content)
-        gr.Markdown("📌 Upload an image containing celebrities, the system will recognize them and provide Wikipedia-based Q&A using the RC-Qwen2-VL model.")
+        gr.Markdown("📌 Upload an image containing celebrities, the system will recognize them and provide Wikipedia-based VQA using the RC-Qwen2-VL model.")
         
         with gr.Row():
             with gr.Column(scale=1):
@@ -450,7 +439,7 @@ def create_interface():
                     label="Confidence Threshold (%)", 
                     info="Adjust the minimum confidence level for celebrity recognition"
                 )
-                submit_button = gr.Button("Ask RC-Qwen2-VL")
+                submit_button = gr.Button("Ask RC-Qwen2VL-7B")
                 
                 # Add examples section
                 if existing_examples:
@@ -512,9 +501,6 @@ def create_interface():
            - "What is the relationship between these people?"
         3. Adjust the confidence threshold slider if needed (lower values will recognize more faces but might be less accurate)
         4. Click the submit button to get the answer
-        5. Or try one of the examples below
-        6. The system caches recognition results for each image and confidence threshold combination
-        7. Cache is automatically cleared when you upload a new image
         """)
     
     return app
@@ -522,6 +508,4 @@ def create_interface():
 # Launch the application
 if __name__ == "__main__":
     app = create_interface()
-    app.launch(share=False)
-
-#bash /mnt/14T-disk/code/instance-detection/run_script.sh
+    app.launch(share=True)