Update app.py

app.py

@@ -17,25 +17,20 @@ try:
 except KeyError:
     raise gr.Error("FATAL: GEMINI_API_KEY not found. Please set it in your Hugging Face Space secrets.")
 
-# --- Main Gradio Function ---
+# --- Main Gradio Generator Function ---
 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
+        # Yield a single update and stop
+        yield "Please upload a video first.", "", "", None, None
+        return
 
-    # --- DINO Analysis (runs in parallel with Gemini upload) ---
-    print("--- Starting DINO Analysis ---")
-    try:
-        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 f"Error in DINO processing: {e}", "", "", None, None
+    # --- Stage 1: Gemini Analysis (Fast) ---
+    print("--- Starting Device Analysis ---")
+    device_type, condition, reason = "Processing...", "Processing...", "Processing..."
+    
+    # Immediately update the UI to show that Gemini is working
+    yield device_type, condition, reason, None, None
 
-    # --- Gemini Analysis ---
-    print("--- Starting Gemini Analysis ---")
     uploaded_file = None
     try:
         uploaded_file = client.files.upload(file=video_file_path)
@@ -45,42 +40,22 @@ def full_analysis(video_file_path, user_observations):
         if uploaded_file.state.name != "ACTIVE":
             raise Exception(f"File processing failed for Gemini. State: {uploaded_file.state.name}")
         
-        # --- 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"
-        }}
+        1. Identify the primary object. Is it an electronic device or a white good?
+        2. If YES, respond with a JSON object with keys: "device_type", "condition" (choose from "unusable", "poor", "functional", "good", "excellent"), and "reason" (a brief explanation).
+        3. If NO, respond with a JSON object with only ONE key: "device_type" starting with "Invalid: " (e.g., "Invalid: Water Bottle").
         """
         
-        model_name = "gemini-1.5-flash-latest"
+        model_name = "gemini-2.5-flash"
         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")
         if "Invalid:" in device_type:
             condition = "N/A"
@@ -89,7 +64,7 @@ def full_analysis(video_file_path, user_observations):
             condition = parsed_json.get("condition", "N/A")
             reason = parsed_json.get("reason", "N/A")
             
-        print("--- Gemini Analysis Complete ---")
+        print("--- Device Analysis Complete ---")
         
     except Exception as e:
         print(f"ERROR during Gemini processing: {e}")
@@ -98,46 +73,58 @@ def full_analysis(video_file_path, user_observations):
         if uploaded_file:
             client.files.delete(name=uploaded_file.name)
 
-    # Return all the results to the Gradio UI
-    return device_type, condition, reason, overlay_images, attention_maps
+    # --- YIELD 1: Update UI with Gemini results ---
+    # The DINO galleries are still empty (None) at this point.
+    yield device_type, condition, reason, None, None
+
+    # --- Stage 2: DINO Analysis (Slow) ---
+    print("--- Starting DINO Analysis ---")
+    try:
+        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}")
+        # If DINO fails, we still want to keep the Gemini results.
+        # We can update one of the text fields to show the DINO error.
+        reason += f" | DINO Error: {e}"
+        overlay_images, attention_maps = None, None
+
+    # --- YIELD 2: Final update with all results ---
+    # This updates the galleries with the DINO images.
+    yield device_type, condition, reason, overlay_images, attention_maps
 
 # --- Gradio Interface ---
-# Define a custom theme
 theme = gr.themes.Soft(
-    primary_hue="green",
-    secondary_hue="emerald",
+    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",
+    body_background_fill="#E0F2F1", 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.")
     
     with gr.Row():
         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..."
+                placeholder="e.g., Device overheats, screen flickers..."
             )
             submit_button = gr.Button("Run Full Analysis", variant="primary")
         
         with gr.Column(scale=2):
-            gr.Markdown("## Gemini Condition Analysis")
+            gr.Markdown("## Device 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")
+                device_type_output = gr.Textbox(label="Device Type", interactive=False)
+                condition_output = gr.Textbox(label="Condition", interactive=False)
+                reason_output = gr.Textbox(label="Reason / Details", interactive=False)
     
-    gr.Markdown("---") # Adds a horizontal line for separation
+    gr.Markdown("---")
     
     gr.Markdown("## DINO: Visual Feature Analysis")
     with gr.Row():
@@ -148,9 +135,10 @@ with gr.Blocks(theme=theme) as demo:
             gr.Markdown("### Self-Attention Maps")
             attention_gallery = gr.Gallery(label="Self-Attention Maps", elem_id="gallery", columns=5)
 
+    # The .click() event now points to a generator function.
+    # Gradio will automatically handle the multiple `yield` statements.
     submit_button.click(
         fn=full_analysis,
-        # Add the new observations_input to the inputs list
         inputs=[video_input, observations_input],
         outputs=[
             device_type_output,