Upload 121 files

app.py

@@ -65,14 +65,12 @@ def gpu_wrapped_execute_video(
     input_video_path,
     flag_relative_input,
     flag_do_crop_input,
-    flag_remap_input,
 ):
     return gradio_pipeline.execute_video(
         input_image_path,
         input_video_path,
         flag_relative_input,
         flag_do_crop_input,
-        flag_remap_input,
     )
 
 @maybe_gpu
@@ -96,11 +94,11 @@ title_md = "assets/gradio_title.md"
 example_portrait_dir = "assets/examples/source"
 example_video_dir = "assets/examples/driving"
 data_examples = [
-    [osp.join(example_portrait_dir, "s9.jpg"), osp.join(example_video_dir, "d0.mp4"), True, True, True],
-    [osp.join(example_portrait_dir, "s6.jpg"), osp.join(example_video_dir, "d0.mp4"), True, True, True],
-    [osp.join(example_portrait_dir, "s10.jpg"), osp.join(example_video_dir, "d5.mp4"), True, True, True],
-    [osp.join(example_portrait_dir, "s5.jpg"), osp.join(example_video_dir, "d6.mp4"), True, True, True],
-    [osp.join(example_portrait_dir, "s7.jpg"), osp.join(example_video_dir, "d7.mp4"), True, True, True],
+    [osp.join(example_portrait_dir, "s9.jpg"), osp.join(example_video_dir, "d0.mp4"), True, True],
+    [osp.join(example_portrait_dir, "s6.jpg"), osp.join(example_video_dir, "d0.mp4"), True, True],
+    [osp.join(example_portrait_dir, "s10.jpg"), osp.join(example_video_dir, "d5.mp4"), True, True],
+    [osp.join(example_portrait_dir, "s5.jpg"), osp.join(example_video_dir, "d6.mp4"), True, True],
+    [osp.join(example_portrait_dir, "s7.jpg"), osp.join(example_video_dir, "d7.mp4"), True, True],
 ]
 #################### interface logic ####################
 
@@ -111,7 +109,6 @@ retargeting_input_image = gr.Image(type="numpy")
 output_image = gr.Image(type="numpy")
 output_image_paste_back = gr.Image(type="numpy")
 output_video = gr.Video()
-output_video_concat = gr.Video()
 
 with gr.Blocks(theme=gr.themes.Soft(),title="LivePortrait: Revolutionary AI-Powered Portrait Animation Technology") as demo:
     gr.HTML(load_description(title_md))
@@ -148,19 +145,15 @@ with gr.Blocks(theme=gr.themes.Soft(),title="LivePortrait: Revolutionary AI-Powe
             with gr.Row():
                 flag_relative_input = gr.Checkbox(value=True, label="relative motion")
                 flag_do_crop_input = gr.Checkbox(value=True, label="do crop")
-                flag_remap_input = gr.Checkbox(value=True, label="paste-back")
     with gr.Row():
         with gr.Column():
             process_button_animation = gr.Button("🚀 Animate", variant="primary")
         with gr.Column():
-            process_button_reset = gr.ClearButton([image_input, video_input, output_video, output_video_concat], value="🧹 Clear")
+            process_button_reset = gr.ClearButton([image_input, video_input, output_video], value="🧹 Clear")
     with gr.Row():
         with gr.Column():
-            with gr.Accordion(open=True, label="The animated video in the original image space"):
+            with gr.Accordion(open=True, label="Avatar animation video"):
                 output_video.render()
-        with gr.Column():
-            with gr.Accordion(open=True, label="The animated video"):
-                output_video_concat.render()
     with gr.Row():
         # Examples
         gr.Markdown("## You could choose the examples below ⬇️")
@@ -173,9 +166,8 @@ with gr.Blocks(theme=gr.themes.Soft(),title="LivePortrait: Revolutionary AI-Powe
                 video_input,
                 flag_relative_input,
                 flag_do_crop_input,
-                flag_remap_input
             ],
-            outputs=[output_video, output_video_concat],
+            outputs=[output_video],
             examples_per_page=5,
             cache_examples=False,
         )
@@ -220,9 +212,8 @@ with gr.Blocks(theme=gr.themes.Soft(),title="LivePortrait: Revolutionary AI-Powe
             video_input,
             flag_relative_input,
             flag_do_crop_input,
-            flag_remap_input
         ],
-        outputs=[output_video, output_video_concat],
+        outputs=[output_video],
         show_progress=True
     )
     image_input.change(

src/config/argument_config.py

@@ -26,7 +26,6 @@ class ArgumentConfig(PrintableConfig):
     flag_lip_retargeting: bool = False
     flag_stitching: bool = True  # we recommend setting it to True!
     flag_relative: bool = True  # whether to use relative motion
-    flag_pasteback: bool = True  # whether to paste-back/stitch the animated face cropping from the face-cropping space to the original image space
     flag_do_crop: bool = True  # whether to crop the source portrait to the face-cropping space
     flag_do_rot: bool = True  # whether to conduct the rotation when flag_do_crop is True
     #########################################

src/config/inference_config.py

@@ -39,7 +39,6 @@ class InferenceConfig(PrintableConfig):
     crf: int = 15  # crf for output video
 
     flag_write_result: bool = True  # whether to write output video
-    flag_pasteback: bool = True  # whether to paste-back/stitch the animated face cropping from the face-cropping space to the original image space
     mask_crop: ndarray = cv2.imread(make_abs_path('../utils/resources/mask_template.png'), cv2.IMREAD_COLOR)
     flag_write_gif: bool = False
     size_gif: int = 256

src/gradio_pipeline.py

@@ -71,7 +71,6 @@ class GradioPipeline(LivePortraitPipeline):
         input_video_path,
         flag_relative_input,
         flag_do_crop_input,
-        flag_remap_input,
         ):
         """ for video driven potrait animation
         """
@@ -82,16 +81,15 @@ class GradioPipeline(LivePortraitPipeline):
                 'driving_info': input_video_path,
                 'flag_relative': flag_relative_input,
                 'flag_do_crop': flag_do_crop_input,
-                'flag_pasteback': flag_remap_input,
             }
             # update config from user input
             self.args = update_args(self.args, args_user)
             self.live_portrait_wrapper.update_config(self.args.__dict__)
             self.cropper.update_config(self.args.__dict__)
             # video driven animation
-            video_path, video_path_concat = self.execute(self.args)
+            video_path = self.execute(self.args)
             # gr.Info("Run successfully!", duration=2)
-            return video_path, video_path_concat,
+            return video_path
         else:
             raise gr.Error("The input source portrait or driving video hasn't been prepared yet 💥!", duration=5)
 

src/live_portrait_pipeline.py

@@ -19,8 +19,8 @@ from .config.inference_config import InferenceConfig
 from .config.crop_config import CropConfig
 from .utils.cropper import Cropper
 from .utils.camera import get_rotation_matrix
-from .utils.video import images2video, concat_frames
-from .utils.crop import _transform_img, prepare_paste_back, paste_back
+from .utils.video import images2video
+from .utils.crop import _transform_img
 from .utils.retargeting_utils import calc_lip_close_ratio
 from .utils.io import load_image_rgb, load_driving_info, resize_to_limit
 from .utils.helper import mkdir, basename, dct2device, is_video, is_template
@@ -46,7 +46,7 @@ class LivePortraitPipeline(object):
         log(f"Load source image from {args.source_image}")
         crop_info = self.cropper.crop_single_image(img_rgb)
         source_lmk = crop_info['lmk_crop']
-        img_crop, img_crop_256x256 = crop_info['img_crop'], crop_info['img_crop_256x256']
+        img_crop_256x256 = crop_info['img_crop_256x256']
         if inference_cfg.flag_do_crop:
             I_s = self.live_portrait_wrapper.prepare_source(img_crop_256x256)
         else:
@@ -88,12 +88,6 @@ class LivePortraitPipeline(object):
             raise Exception("Unsupported driving types!")
         #########################################
 
-        ######## prepare for pasteback ########
-        if inference_cfg.flag_pasteback:
-            mask_ori = prepare_paste_back(inference_cfg.mask_crop, crop_info['M_c2o'], dsize=(img_rgb.shape[1], img_rgb.shape[0]))
-            I_p_paste_lst = []
-        #########################################
-
         I_p_lst = []
         R_d_0, x_d_0_info = None, None
         for i in track(range(n_frames), description='Animating...', total=n_frames):
@@ -168,23 +162,9 @@ class LivePortraitPipeline(object):
             I_p_i = self.live_portrait_wrapper.parse_output(out['out'])[0]
             I_p_lst.append(I_p_i)
 
-            if inference_cfg.flag_pasteback:
-                I_p_i_to_ori_blend = paste_back(I_p_i, crop_info['M_c2o'], img_rgb, mask_ori)
-                I_p_paste_lst.append(I_p_i_to_ori_blend)
-
         mkdir(args.output_dir)
-        wfp_concat = None
-        if is_video(args.driving_info):
-            frames_concatenated = concat_frames(I_p_lst, driving_rgb_lst, img_crop_256x256)
-            # save (driving frames, source image, drived frames) result
-            wfp_concat = osp.join(args.output_dir, f'{basename(args.source_image)}--{basename(args.driving_info)}_concat.mp4')
-            images2video(frames_concatenated, wfp=wfp_concat)
-
-        # save drived result
+        # save avatar-only animation result
         wfp = osp.join(args.output_dir, f'{basename(args.source_image)}--{basename(args.driving_info)}.mp4')
-        if inference_cfg.flag_pasteback:
-            images2video(I_p_paste_lst, wfp=wfp)
-        else:
-            images2video(I_p_lst, wfp=wfp)
+        images2video(I_p_lst, wfp=wfp)
 
-        return wfp, wfp_concat
+        return wfp