Update app.py

app.py

@@ -415,18 +415,39 @@ def analyze_video_sam3(video_path, text_prompts, hf_token, max_duration=8.0, pro
             
     progress(1.0, desc="Done!")
     
+    # Calculate Max-Visible Count per concept (maximum visible in any single keyframe)
+    max_visible = defaultdict(int)
+    for k_idx in range(len(keyframes)):
+        frame_masks = masks_by_frame.get(k_idx, {})
+        current_visible = defaultdict(int)
+        for obj_id, mask in frame_masks.items():
+            if np.any(mask):
+                prompt = obj_to_prompt.get(obj_id)
+                if prompt:
+                    current_visible[prompt] += 1
+        for prompt, count in current_visible.items():
+            if count > max_visible[prompt]:
+                max_visible[prompt] = count
+                
     # Generate list report
-    report_md = "# 📋 Tracked Object ID Details\n\n"
-    report_md += "Below is the list of all tracked visual instances associated with each text concept:\n\n"
+    report_md = "# 📋 Apartment Inspection & Counting Report\n\n"
+    report_md += "Below is the comparison between the **Total Tracked IDs** (which can double-count objects if they temporarily go out of view and get assigned a new ID) and the **Max-Visible Count** (which avoids double-counting by finding the peak number of objects visible simultaneously in a single frame):\n\n"
+    
+    # Create a markdown table
+    report_md += "| Category | Unique Tracked IDs | Max-Visible Count | Detailed Tracking IDs |\n"
+    report_md += "| :--- | :---: | :---: | :--- |\n"
     
     for prompt in sorted(track_db.keys()):
         obj_ids = sorted(list(track_db[prompt]))
         if obj_ids:
             ids_str = ", ".join(f"`#{oid}`" for oid in obj_ids)
-            report_md += f"- **{prompt.capitalize()}**: {len(obj_ids)} items (IDs: {ids_str})\n"
+            tracked_count = len(obj_ids)
+            max_vis = max_visible.get(prompt, 0)
+            category_name = prompt.capitalize()
+            report_md += f"| **{category_name}** | {tracked_count} | **{max_vis}** | {ids_str} |\n"
             
     if not track_db:
-        report_md += "*No objects detected in the video for the given prompts.*"
+        report_md += "\n*No objects detected in the video for the given prompts.*"
         
     return final_output_path, report_md