Update module

app.py

@@ -88,10 +88,13 @@ def run_avse_inference(video_path, audio_path):
     noisy = wavfile.read(audio_path)[1].astype(np.float32) / (2 ** 15)
 
     # Load grayscale video
-    vr = VideoReader(video_path, ctx=cpu(0))
+    vr = VideoReader(output_path, ctx=cpu(0))
     frames = vr.get_batch(list(range(len(vr)))).asnumpy()
-    bg_frames = np.array([cv2.cvtColor(f, cv2.COLOR_RGB2GRAY) for f in frames]).astype(np.float32) / 255.0
-    bg_frames = np.expand_dims(bg_frames, axis=0)
+    bg_frames = np.array([
+        cv2.cvtColor(frames[i], cv2.COLOR_RGB2GRAY) for i in range(len(frames))
+    ]).astype(np.float32)
+    bg_frames /= 255.0
+
 
     # Combine into input dict (match what model.enhance expects)
     data = {
@@ -192,13 +195,6 @@ def extract_faces(video_file):
     # ------------------------------- #
     # AVSE models
 
-    vr = VideoReader(output_path, ctx=cpu(0))
-    frames = vr.get_batch(list(range(len(vr)))).asnumpy()
-    bg_frames = np.array([
-        cv2.cvtColor(frames[i], cv2.COLOR_RGB2GRAY) for i in range(len(frames))
-    ]).astype(np.float32)
-    bg_frames /= 255.0
-
     enhanced_audio_path = run_avse_inference(output_path, audio_path)