File size: 7,160 Bytes
a891a57
 
 
 
 
 
 
 
 
 
 
 
 
 
61ea780
a891a57
4f1874e
d4a5c81
 
a891a57
 
 
 
 
 
 
 
 
 
 
d55e5c1
a891a57
 
 
 
 
913899c
a891a57
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
fef0beb
 
c687a76
 
 
 
 
fef0beb
 
 
 
a891a57
 
9823588
d3cb9a3
61ea780
d3cb9a3
 
 
 
b8d159a
d3cb9a3
9823588
a891a57
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c687a76
a891a57
 
 
 
 
 
 
c687a76
15ec329
 
a891a57
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d55e5c1
a891a57
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d4a5c81
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
# coding: utf-8

"""
The entrance of the gradio
"""

import tyro
import gradio as gr
import os.path as osp
from src.utils.helper import load_description
from src.gradio_pipeline import GradioPipeline
from src.config.crop_config import CropConfig
from src.config.argument_config import ArgumentConfig
from src.config.inference_config import InferenceConfig
import spaces

# import gdown
# folder_url = f"https://drive.google.com/drive/folders/1UtKgzKjFAOmZkhNK-OYT0caJ_w2XAnib"
# gdown.download_folder(url=folder_url, output="pretrained_weights", quiet=False)

def partial_fields(target_class, kwargs):
    return target_class(**{k: v for k, v in kwargs.items() if hasattr(target_class, k)})

# set tyro theme
tyro.extras.set_accent_color("bright_cyan")
args = tyro.cli(ArgumentConfig)

# specify configs for inference
inference_cfg = partial_fields(InferenceConfig, args.__dict__)  # use attribute of args to initial InferenceConfig
crop_cfg = partial_fields(CropConfig, args.__dict__)  # use attribute of args to initial CropConfig

gradio_pipeline = GradioPipeline(
    inference_cfg=inference_cfg,
    crop_cfg=crop_cfg,
    args=args
)

# assets
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, True],
    [osp.join(example_portrait_dir, "s6.jpg"), osp.join(example_video_dir, "d0.mp4"), True, True, True, True],
    [osp.join(example_portrait_dir, "s10.jpg"), osp.join(example_video_dir, "d5.mp4"), True, True, True, True],
    [osp.join(example_portrait_dir, "s5.jpg"), osp.join(example_video_dir, "d6.mp4"), True, True, True, True],
    [osp.join(example_portrait_dir, "s7.jpg"), osp.join(example_video_dir, "d7.mp4"), True, True, True, True],
]
#################### interface logic ####################

# Define components first
eye_retargeting_slider = gr.Slider(minimum=0, maximum=0.8, step=0.01, label="target eyes-open ratio")
lip_retargeting_slider = gr.Slider(minimum=0, maximum=0.8, step=0.01, label="target lip-open ratio")
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()) as demo:
    gr.HTML(load_description(title_md))
    gr.Markdown(load_description("assets/gradio_description_upload.md"))
    with gr.Row():
        with gr.Accordion(open=True, label="Source Portrait"):
            image_input = gr.Image(type="filepath")
            gr.Examples(
                examples=[
                    [osp.join(example_portrait_dir, "s9.jpg")],
                    [osp.join(example_portrait_dir, "s6.jpg")],
                    [osp.join(example_portrait_dir, "s10.jpg")],
                    [osp.join(example_portrait_dir, "s5.jpg")],
                    [osp.join(example_portrait_dir, "s7.jpg")],
                ],
                inputs=[image_input],
                cache_examples=False,
            )
        with gr.Accordion(open=True, label="Driving Video"):
            video_input = gr.Video()
            gr.Examples(
                examples=[
                    [osp.join(example_video_dir, "d0.mp4")],
                    [osp.join(example_video_dir, "d5.mp4")],
                    [osp.join(example_video_dir, "d6.mp4")],
                    [osp.join(example_video_dir, "d7.mp4")],
                ],
                inputs=[video_input],
                cache_examples=False,
            )
    gr.Markdown(load_description("assets/gradio_description_animation.md"))
    with gr.Row():
        with gr.Accordion(open=True, label="Animation Options"):
            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")
    with gr.Row():
        with gr.Column():
            with gr.Accordion(open=True, label="The animated video in the original image space"):
                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 ⬇️")
    with gr.Row():
        gr.Examples(
            examples=data_examples,
            fn=lambda *args: spaces.GPU()(gradio_pipeline.execute_video)(*args),
            inputs=[
                image_input,
                video_input,
                flag_relative_input,
                flag_do_crop_input,
                flag_remap_input
            ],
            outputs=[output_image, output_image_paste_back],
            examples_per_page=5,
            cache_examples="lazy",
        )
    gr.Markdown(load_description("assets/gradio_description_retargeting.md"))
    with gr.Row():
        eye_retargeting_slider.render()
        lip_retargeting_slider.render()
    with gr.Row():
        process_button_retargeting = gr.Button("🚗 Retargeting", variant="primary")
        process_button_reset_retargeting = gr.ClearButton(
            [
                eye_retargeting_slider,
                lip_retargeting_slider,
                retargeting_input_image,
                output_image,
                output_image_paste_back
            ],
            value="🧹 Clear"
        )
    with gr.Row():
        with gr.Column():
            with gr.Accordion(open=True, label="Retargeting Input"):
                retargeting_input_image.render()
        with gr.Column():
            with gr.Accordion(open=True, label="Retargeting Result"):
                output_image.render()
        with gr.Column():
            with gr.Accordion(open=True, label="Paste-back Result"):
                output_image_paste_back.render()
    # binding functions for buttons
    process_button_retargeting.click(
        fn=gradio_pipeline.execute_image,
        inputs=[eye_retargeting_slider, lip_retargeting_slider],
        outputs=[output_image, output_image_paste_back],
        show_progress=True
    )
    process_button_animation.click(
        fn=lambda *args: spaces.GPU()(gradio_pipeline.execute_video)(*args),
        inputs=[
            image_input,
            video_input,
            flag_relative_input,
            flag_do_crop_input,
            flag_remap_input
        ],
        outputs=[output_video, output_video_concat],
        show_progress=True
    )
    image_input.change(
        fn=gradio_pipeline.prepare_retargeting,
        inputs=image_input,
        outputs=[eye_retargeting_slider, lip_retargeting_slider, retargeting_input_image]
    )

demo.launch()