debug

app.py

@@ -93,12 +93,12 @@ pipe = Pose2VideoPipeline(
 )
 pipe = pipe.to("cuda", dtype=weight_dtype)
 
-lmk_extractor = LMKExtractor()
-vis = FaceMeshVisualizer()
+# lmk_extractor = LMKExtractor()
+# vis = FaceMeshVisualizer()
 
 frame_inter_model = init_frame_interpolation_model()
 
-@spaces.GPU  # (duration=100)
+@spaces.GPU
 def audio2video(input_audio, ref_img, headpose_video=None, size=512, steps=25, length=60, seed=42):
     print('=====Start processing======')
     
@@ -107,6 +107,9 @@ def audio2video(input_audio, ref_img, headpose_video=None, size=512, steps=25, l
     fi_step = 3
 
     generator = torch.manual_seed(seed)
+    
+    lmk_extractor = LMKExtractor()
+    vis = FaceMeshVisualizer()
 
     width, height = size, size    
 
@@ -226,7 +229,7 @@ def audio2video(input_audio, ref_img, headpose_video=None, size=512, steps=25, l
     
     return save_path.replace('_noaudio.mp4', '.mp4'), ref_image_pil
 
-@spaces.GPU  # (duration=100)
+@spaces.GPU
 def video2video(ref_img, source_video, size=512, steps=25, length=60, seed=42):
     print('=====Start processing======')
     
@@ -234,6 +237,9 @@ def video2video(ref_img, source_video, size=512, steps=25, length=60, seed=42):
     fi_step = 3
     
     generator = torch.manual_seed(seed)
+    
+    lmk_extractor = LMKExtractor()
+    vis = FaceMeshVisualizer()
 
     width, height = size, size
 
@@ -247,7 +253,7 @@ def video2video(ref_img, source_video, size=512, steps=25, length=60, seed=42):
     save_dir.mkdir(exist_ok=True, parents=True)
     
     ref_image_np = cv2.cvtColor(ref_img, cv2.COLOR_RGB2BGR)
-    ref_image_np = crop_face(ref_image_np, lmk_extractor)
+    # ref_image_np = crop_face(ref_image_np, lmk_extractor)
     if ref_image_np is None:
         return None, Image.fromarray(ref_img)