Update difpoint/inference.py

difpoint/inference.py

@@ -169,7 +169,7 @@ class Inferencer(object):
         self.point_diffusion.to(self.device)
 
         self.lm_croper_checkpoint = './downloaded_repo/ckpts/shape_predictor_68_face_landmarks.dat'
-        #self.croper = Croper(lm_croper_checkpoint)
+        self.croper = Croper(lm_croper_checkpoint)
 
         self.norm_info = dict(np.load(r'difpoint/datasets/norm_info_d6.5_c8.5_vox1_train.npz'))
 
@@ -182,7 +182,7 @@ class Inferencer(object):
         self.inf_cfg = OmegaConf.load("difpoint/configs/onnx_infer.yaml")
         self.inf_cfg.infer_params.flag_pasteback = False
 
-        #self.live_portrait_pipeline = FasterLivePortraitPipeline(cfg=self.inf_cfg, is_animal=False)
+        self.live_portrait_pipeline = FasterLivePortraitPipeline(cfg=self.inf_cfg, is_animal=False)
         #ret = self.live_portrait_pipeline.prepare_source(source_image)
 
         print('#'*25+f'End initialization, cost time {time.time()-st}'+'#'*25)
@@ -273,9 +273,7 @@ class Inferencer(object):
     # 2024.06.26
     @torch.no_grad()
     def generate_with_audio_img(self, upload_audio_path, tts_audio_path, audio_type, image_path, smoothed_pitch, smoothed_yaw, smoothed_roll, smoothed_t,  save_path='results'):
-        self.live_portrait_pipeline = FasterLivePortraitPipeline(cfg=self.inf_cfg, is_animal=False)
-        self.croper = Croper(self.lm_croper_checkpoint)
-        self.live_portrait_pipeline.init()
+        
         if audio_type == 'upload':
             audio_path = upload_audio_path
         elif audio_type == 'tts':