Update videoretalking/inference_function.py

videoretalking/inference_function.py

@@ -14,8 +14,8 @@ from third_part.face3d.util.load_mats import load_lm3d
 from third_part.face3d.extract_kp_videos import KeypointExtractor
 # face enhancement
 from third_part.GPEN.gpen_face_enhancer import FaceEnhancement
-# expression control
-from third_part.ganimation_replicate.model.ganimation import GANimationModel
+# # expression control
+# from third_part.ganimation_replicate.model.ganimation import GANimationModel
 
 from utils import audio
 from utils.ffhq_preprocess import Croper
@@ -206,10 +206,10 @@ def video_lipsync_correctness(face, audio_path, outfile=None, tmp_dir="temp", cr
     frame_h, frame_w = full_frames[0].shape[:-1]
     out = cv2.VideoWriter('temp/{}/result.mp4'.format(tmp_dir), cv2.VideoWriter_fourcc(*'mp4v'), fps, (frame_w, frame_h))
     
-    if up_face != 'original':
-        instance = GANimationModel()
-        instance.initialize()
-        instance.setup()
+    # if up_face != 'original':
+    #     instance = GANimationModel()
+    #     instance.initialize()
+    #     instance.setup()
     
     kp_extractor = KeypointExtractor()
     for i, (img_batch, mel_batch, frames, coords, img_original, f_frames) in enumerate(tqdm(gen, desc='[Step 6] Lip Synthesis:', total=int(np.ceil(float(len(mel_chunks)) / LNet_batch_size)))):
@@ -229,12 +229,12 @@ def video_lipsync_correctness(face, audio_path, outfile=None, tmp_dir="temp", cr
             
             if up_face == 'original':
                 cur_gen_faces = img_original
-            else:
-                test_batch = {'src_img': torch.nn.functional.interpolate((img_original * 2 - 1), size=(128, 128), mode='bilinear'), 
-                              'tar_aus': tar_aus.repeat(len(incomplete), 1)}
-                instance.feed_batch(test_batch)
-                instance.forward()
-                cur_gen_faces = torch.nn.functional.interpolate(instance.fake_img / 2. + 0.5, size=(384, 384), mode='bilinear')
+            # else:
+            #     test_batch = {'src_img': torch.nn.functional.interpolate((img_original * 2 - 1), size=(128, 128), mode='bilinear'), 
+            #                   'tar_aus': tar_aus.repeat(len(incomplete), 1)}
+            #     instance.feed_batch(test_batch)
+            #     instance.forward()
+            #     cur_gen_faces = torch.nn.functional.interpolate(instance.fake_img / 2. + 0.5, size=(384, 384), mode='bilinear')
                 
             if without_rl1 is not False:
                 incomplete, reference = torch.split(img_batch, 3, dim=1)