v0.0.2.beta

app.py

@@ -25,6 +25,12 @@ def sadtalker_demo():
                     <a style='font-size:18px;color: #efefef' href='https://sadtalker.github.io'>Homepage</a>  &nbsp;&nbsp;&nbsp;&nbsp;&nbsp; \
                      <a style='font-size:18px;color: #efefef' href='https://github.com/Winfredy/SadTalker'> Github </div>")
         
+        
+        gr.Markdown("""
+        <b>You may duplicate the space and upgrade to GPU in settings for better performance and faster inference without waiting in the queue. <a style='display:inline-block' href="https://huggingface.co/spaces/vinthony/SadTalker?duplicate=true"><img src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a></b> \
+        <br/><b>Alternatively, try our GitHub <a href=https://github.com/Winfredy/SadTalker> code </a> on your own GPU. </b> <a style='display:inline-block' href="https://github.com/Winfredy/SadTalker"><img src="https://img.shields.io/github/stars/Winfredy/SadTalker?style=social"/></a> \
+        """)
+        
         with gr.Row().style(equal_height=False):
             with gr.Column(variant='panel'):
                 with gr.Tabs(elem_id="sadtalker_source_image"):
@@ -130,7 +136,7 @@ def sadtalker_demo():
                             enhancer], 
                         outputs=[gen_video],
                         fn=sad_talker.test,
-                        cache_examples=False) # os.getenv('SYSTEM') == 'spaces')
+                        cache_examples=os.getenv('SYSTEM') == 'spaces') # 
 
         submit.click(
                     fn=sad_talker.test, 
@@ -148,6 +154,7 @@ def sadtalker_demo():
 if __name__ == "__main__":
 
     demo = sadtalker_demo()
-    demo.launch()
+    demo.queue(max_size=10)
+    demo.launch(debug=True)
 
 

src/face3d/extract_kp_videos.py

@@ -13,7 +13,7 @@ from torch.multiprocessing import Pool, Process, set_start_method
 
 class KeypointExtractor():
     def __init__(self, device):
-        self.detector = face_alignment.FaceAlignment(face_alignment.LandmarksType._2D, device='cpu')   
+        self.detector = face_alignment.FaceAlignment(face_alignment.LandmarksType._2D, device=device)   
 
     def extract_keypoint(self, images, name=None, info=True):
         if isinstance(images, list):

src/gradio_demo.py

@@ -42,8 +42,6 @@ class SadTalker():
 
         self.free_view_checkpoint = os.path.join( checkpoint_path, 'facevid2vid_00189-model.pth.tar')
 
-       
-        
         self.lazy_load = lazy_load
 
         if not self.lazy_load:
@@ -75,6 +73,7 @@ class SadTalker():
             self.mapping_checkpoint = os.path.join(self.checkpoint_path, 'mapping_00229-model.pth.tar')
             self.facerender_yaml_path = os.path.join(self.config_path, 'facerender.yaml')
 
+        print(self.mapping_checkpoint)
         print(self.free_view_checkpoint)
         self.animate_from_coeff = AnimateFromCoeff(self.free_view_checkpoint, self.mapping_checkpoint, 
                                             self.facerender_yaml_path, self.device)
@@ -117,7 +116,7 @@ class SadTalker():
         batch = get_data(first_coeff_path, audio_path, self.device, ref_eyeblink_coeff_path=None, still=still_mode) # longer audio?
         coeff_path = self.audio_to_coeff.generate(batch, save_dir, pose_style)
         #coeff2video
-        batch_size = 2
+        batch_size = 8
         data = get_facerender_data(coeff_path, crop_pic_path, first_coeff_path, audio_path, batch_size, still_mode=still_mode, preprocess=preprocess)
         return_path = self.animate_from_coeff.generate(data, save_dir,  pic_path, crop_info, enhancer='gfpgan' if use_enhancer else None, preprocess=preprocess)
         video_name = data['video_name']