add debug

app.py

@@ -24,6 +24,7 @@ class ChaplinGradio:
         self.frame_buffer = []
         self.min_frames = 32  # 2 seconds of video at 16 fps
         self.last_prediction = ""
+        print(f"Initialized with device: {self.device}, fps: {self.fps}, min_frames: {self.min_frames}")
 
     def download_models(self):
         """Download required model files from HuggingFace"""
@@ -71,26 +72,34 @@ class ChaplinGradio:
         self.last_frame_time = current_time
         
         if frame is None:
+            print("Received None frame")
             return "No video input detected"
         
         try:
+            print(f"Received frame with shape: {frame.shape}")
+            
             # Convert frame to grayscale if it's not already
             if len(frame.shape) == 3:
                 frame = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY)
+                print("Converted frame to grayscale")
             
             # Add frame to buffer
             self.frame_buffer.append(frame)
+            print(f"Buffer size now: {len(self.frame_buffer)}/{self.min_frames}")
             
             # Process when we have enough frames
             if len(self.frame_buffer) >= self.min_frames:
+                print("Processing buffer - have enough frames")
                 # Create temp directory if it doesn't exist
                 os.makedirs("temp", exist_ok=True)
                 
                 # Generate temporary video file path
                 temp_video = f"temp/frames_{time.time_ns()}.mp4"
+                print(f"Created temp video path: {temp_video}")
                 
                 # Get frame dimensions from first frame
                 frame_height, frame_width = self.frame_buffer[0].shape[:2]
+                print(f"Video dimensions: {frame_width}x{frame_height}")
                 
                 # Create video writer
                 out = cv2.VideoWriter(
@@ -102,16 +111,20 @@ class ChaplinGradio:
                 )
                 
                 # Write all frames to video
-                for f in self.frame_buffer:
+                for i, f in enumerate(self.frame_buffer):
                     out.write(f)
+                print(f"Wrote {i+1} frames to video")
                 out.release()
                 
                 # Clear buffer but keep last few frames for continuity
                 self.frame_buffer = self.frame_buffer[-8:]  # Keep last 0.5 seconds
+                print(f"Cleared buffer, kept {len(self.frame_buffer)} frames")
                 
                 try:
                     # Process the video file using the pipeline
+                    print("Starting model inference...")
                     predicted_text = self.vsr_model(temp_video)
+                    print(f"Model prediction: {predicted_text}")
                     if predicted_text:
                         self.last_prediction = predicted_text
                     return self.last_prediction
@@ -123,6 +136,7 @@ class ChaplinGradio:
                     # Clean up temp file
                     if os.path.exists(temp_video):
                         os.remove(temp_video)
+                        print("Cleaned up temp video file")
             
             return self.last_prediction or "Waiting for speech..."
                 
@@ -137,7 +151,10 @@ chaplin = ChaplinGradio()
 iface = gr.Interface(
     fn=chaplin.process_frame,
     inputs=gr.Image(sources=["webcam"], streaming=True),
-    outputs=gr.Textbox(label="Predicted Text", interactive=False),
+    outputs=[
+        gr.Textbox(label="Predicted Text", interactive=False),
+        gr.Textbox(label="Debug Log", interactive=False)
+    ],
     title="Chaplin - Live Visual Speech Recognition",
     description="Speak clearly into the webcam. The model will process your speech in ~2 second chunks.",
     live=True