Update app.py

app.py

@@ -144,16 +144,28 @@ class VideoQAInterface:
         self.frame_index = None
         self.frame_data = None
         self.processed = False
+        self.current_video_path = None  # Store the video path
+        self.temp_dir = tempfile.mkdtemp()
+
+    def __del__(self):
+        """Cleanup temporary files"""
+        if hasattr(self, 'temp_dir') and os.path.exists(self.temp_dir):
+            shutil.rmtree(self.temp_dir, ignore_errors=True)
 
     def process_video(self, video_file, progress=gr.Progress()):
         """Process video with progress tracking"""
         try:
             if video_file is None:
                 return "Please upload a video first."
-                
+            
+            # Save uploaded video to temp directory
+            temp_video_path = os.path.join(self.temp_dir, "input_video.mp4")
+            shutil.copy2(video_file.name, temp_video_path)
+            self.current_video_path = temp_video_path
+            
             progress(0, desc="Starting video processing...")
             self.frame_index, self.frame_data, message = self.processor.process_video(
-                video_file.name, progress
+                self.current_video_path, progress
             )
             
             if self.frame_index is not None:
@@ -170,7 +182,7 @@ class VideoQAInterface:
     @torch.no_grad()
     def answer_question(self, query):
         """Answer questions about the video"""
-        if not self.processed:
+        if not self.processed or self.current_video_path is None:
             return None, "Please process a video first."
             
         try:
@@ -192,22 +204,27 @@ class VideoQAInterface:
             descriptions = []
             frames = []
             
-            cap = cv2.VideoCapture(video_file.name)
-            for result in results:
-                frame_number = result['frame_number']
-                cap.set(cv2.CAP_PROP_POS_FRAMES, frame_number)
-                ret, frame = cap.read()
-                
-                if ret:
-                    frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
-                    frames.append(Image.fromarray(frame_rgb))
+            # Use cv2.VideoCapture to read frames
+            cap = cv2.VideoCapture(self.current_video_path)
+            try:
+                for result in results:
+                    frame_number = result['frame_number']
+                    cap.set(cv2.CAP_PROP_POS_FRAMES, frame_number)
+                    ret, frame = cap.read()
                     
-                    desc = f"Timestamp: {result['timestamp']:.2f}s\n"
-                    desc += f"Scene Description: {result['caption']}\n"
-                    desc += f"Relevance Score: {result['relevance']:.2f}"
-                    descriptions.append(desc)
+                    if ret:
+                        frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+                        frames.append(Image.fromarray(frame_rgb))
+                        
+                        desc = f"Timestamp: {result['timestamp']:.2f}s\n"
+                        desc += f"Scene Description: {result['caption']}\n"
+                        desc += f"Relevance Score: {result['relevance']:.2f}"
+                        descriptions.append(desc)
+            finally:
+                cap.release()  # Ensure video capture is released
             
-            cap.release()
+            if not frames:
+                return None, "No relevant frames found."
             
             combined_desc = "\n\nFrame Analysis:\n\n"
             for i, desc in enumerate(descriptions, 1):