Update app.py

app.py

@@ -7,7 +7,7 @@ from dotenv import load_dotenv
 from google import genai
 from google.genai import types
 
-# Import our new DINO processing function
+# Import our DINO processing function
 from dino_processor import process_video_with_dino
 
 # --- Configuration and Client Initialization ---
@@ -18,29 +18,26 @@ except KeyError:
     raise gr.Error("FATAL: GEMINI_API_KEY not found. Please set it in your Hugging Face Space secrets.")
 
 # --- Main Gradio Function ---
-def full_analysis(video_file_path):
+def full_analysis(video_file_path, user_observations):
     if not video_file_path:
         # Return empty values for all components
         return "Please upload a video first.", "", "", None, None
 
-    # --- DINO Analysis ---
+    # --- DINO Analysis (runs in parallel with Gemini upload) ---
     print("--- Starting DINO Analysis ---")
     try:
-        # This function will return a list of tuples: [(overlay1, attn1), (overlay2, attn2), ...]
         dino_results = process_video_with_dino(video_file_path)
         overlay_images = [res[0] for res in dino_results]
         attention_maps = [res[1] for res in dino_results]
         print("--- DINO Analysis Complete ---")
     except Exception as e:
         print(f"ERROR during DINO processing: {e}")
-        # Return an error message and empty galleries
         return f"Error in DINO processing: {e}", "", "", None, None
 
     # --- Gemini Analysis ---
     print("--- Starting Gemini Analysis ---")
     uploaded_file = None
     try:
-        # Wait for DINO to finish before starting the Gemini upload
         uploaded_file = client.files.upload(file=video_file_path)
         while uploaded_file.state.name == "PROCESSING":
             time.sleep(5)
@@ -48,21 +45,50 @@ def full_analysis(video_file_path):
         if uploaded_file.state.name != "ACTIVE":
             raise Exception(f"File processing failed for Gemini. State: {uploaded_file.state.name}")
         
-        prompt = """
-        Analyze the provided video. Respond ONLY with a valid JSON object with three keys:
-        1. "device_type": A short string identifying the device.
-        2. "condition": A single word: "Mint", "Excellent", "Good", "Fair", or "Poor".
-        3. "reason": A brief string explaining the condition.
+        # --- NEW, MORE ADVANCED PROMPT ---
+        prompt = f"""
+        Analyze the provided video and the user's observations. Respond ONLY with a valid JSON object.
+
+        User Observations: "{user_observations if user_observations else 'None'}"
+
+        Follow these steps:
+        1. First, identify the primary object in the video. Is it an electronic device or a white good (e.g., refrigerator, washing machine, microwave)?
+        2. If it IS an electronic device or white good, respond with a JSON object with these three keys:
+           - "device_type": A short string identifying the specific device (e.g., "iPhone 13 Pro", "Samsung Washing Machine").
+           - "condition": A single word describing its visual condition. Choose ONLY from: "unusable", "poor", "functional", "good", "excellent".
+           - "reason": A very brief, concise string explaining the condition (e.g., "heavy screen cracks", "minor scratches on body", "clean with no visible marks").
+        3. If it is NOT an electronic device or white good, respond with a JSON object with only ONE key:
+           - "device_type": A string starting with "Invalid: " followed by the object's category (e.g., "Invalid: Water Bottle", "Invalid: Shoe").
+
+        Example for a valid device:
+        {{
+          "device_type": "Dell XPS Laptop",
+          "condition": "functional",
+          "reason": "Visible scratches and paint erosion on the lid."
+        }}
+
+        Example for an invalid object:
+        {{
+          "device_type": "Invalid: Coffee Mug"
+        }}
         """
-        model_name = "gemini-1.5-flash-latest" # Using a reliable public model
+        
+        model_name = "gemini-1.5-flash-latest"
         config = types.GenerateContentConfig(temperature=0.2, response_mime_type="application/json")
         contents = [uploaded_file, prompt]
         
         response = client.models.generate_content(model=f"models/{model_name}", contents=contents, config=config)
         parsed_json = json.loads(response.text)
+        
+        # Check if the device was valid or not
         device_type = parsed_json.get("device_type", "N/A")
-        condition = parsed_json.get("condition", "N/A")
-        reason = parsed_json.get("reason", "N/A")
+        if "Invalid:" in device_type:
+            condition = "N/A"
+            reason = "Object is not a valid electronic device or white good."
+        else:
+            condition = parsed_json.get("condition", "N/A")
+            reason = parsed_json.get("reason", "N/A")
+            
         print("--- Gemini Analysis Complete ---")
         
     except Exception as e:
@@ -76,29 +102,56 @@ def full_analysis(video_file_path):
     return device_type, condition, reason, overlay_images, attention_maps
 
 # --- Gradio Interface ---
-with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown("# 📱 Advanced Device Condition Analyzer")
-    gr.Markdown("Upload a video to get a condition analysis from Gemini and attention maps from DINO.")
-    
-    video_input = gr.Video(label="Upload or Record Video", sources=["upload", "webcam"], format="mp4")
-    submit_button = gr.Button("Run Full Analysis", variant="primary")
+# Define a custom theme
+theme = gr.themes.Soft(
+    primary_hue="green",
+    secondary_hue="emerald",
+).set(
+    body_background_fill="#E0F2F1", # A light aquatic green
+    block_background_fill="white",
+    block_border_width="1px",
+    block_shadow="*shadow_drop_lg",
+    button_primary_background_fill="*primary_500",
+    button_primary_text_color="white",
+)
+
+with gr.Blocks(theme=theme) as demo:
+    # Using a refrigerator emoji for white goods 🧊 or 🔌
+    gr.Markdown("# 🧊 KIKERP: AI Assessment Tool")
+    gr.Markdown("Upload a video of a device or white good for a complete visual and AI-driven analysis.")
     
-    gr.Markdown("## Gemini Condition Analysis")
     with gr.Row():
-        device_type_output = gr.Textbox(label="Device Type")
-        condition_output = gr.Textbox(label="Condition")
-        reason_output = gr.Textbox(label="Reason / Details")
+        with gr.Column(scale=1):
+            video_input = gr.Video(label="Upload or Record Video", sources=["upload", "webcam"], format="mp4")
+            # NEW: Optional text input for user observations
+            observations_input = gr.Textbox(
+                label="Optional: User Observations", 
+                placeholder="e.g., Device overheats, screen flickers, makes a strange noise..."
+            )
+            submit_button = gr.Button("Run Full Analysis", variant="primary")
         
-    gr.Markdown("## DINO: Overlayed Heatmaps on Representative Frames")
-    # Use gr.Gallery to display multiple images
-    overlay_gallery = gr.Gallery(label="Overlayed Heatmaps", elem_id="gallery")
+        with gr.Column(scale=2):
+            gr.Markdown("## Gemini Condition Analysis")
+            with gr.Row():
+                device_type_output = gr.Textbox(label="Device Type")
+                condition_output = gr.Textbox(label="Condition")
+                reason_output = gr.Textbox(label="Reason / Details")
     
-    gr.Markdown("## DINO: Self-Attention Maps")
-    attention_gallery = gr.Gallery(label="Self-Attention Maps", elem_id="gallery")
+    gr.Markdown("---") # Adds a horizontal line for separation
+    
+    gr.Markdown("## DINO: Visual Feature Analysis")
+    with gr.Row():
+        with gr.Column():
+            gr.Markdown("### Overlayed Heatmaps on Key Frames")
+            overlay_gallery = gr.Gallery(label="Overlayed Heatmaps", elem_id="gallery", columns=5)
+        with gr.Column():
+            gr.Markdown("### Self-Attention Maps")
+            attention_gallery = gr.Gallery(label="Self-Attention Maps", elem_id="gallery", columns=5)
 
     submit_button.click(
         fn=full_analysis,
-        inputs=video_input,
+        # Add the new observations_input to the inputs list
+        inputs=[video_input, observations_input],
         outputs=[
             device_type_output, 
             condition_output,