File size: 4,843 Bytes
9bdd97c
 
 
 
 
 
 
 
 
 
f57e36f
9bdd97c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f57e36f
 
9bdd97c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f57e36f
 
 
 
9bdd97c
f57e36f
9bdd97c
 
f57e36f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9bdd97c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f57e36f
9bdd97c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#!/usr/bin/env python

from __future__ import annotations

import argparse
import functools
import os
import pathlib
import subprocess
import sys
import tarfile

# workaround for https://github.com/gradio-app/gradio/issues/483
command = 'pip install -U gradio==2.7.0'
subprocess.call(command.split())

import gradio as gr
import huggingface_hub
import PIL.Image
import torch
import torchvision

sys.path.insert(0, 'bizarre-pose-estimator')

from _util.twodee_v0 import I as ImageWrapper

TOKEN = os.environ['TOKEN']

MODEL_REPO = 'hysts/bizarre-pose-estimator-models'
MODEL_FILENAME = 'tagger.pth'
LABEL_FILENAME = 'tags.txt'


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser()
    parser.add_argument('--device', type=str, default='cpu')
    parser.add_argument('--score-slider-step', type=float, default=0.05)
    parser.add_argument('--score-threshold', type=float, default=0.5)
    parser.add_argument('--theme', type=str, default='dark-grass')
    parser.add_argument('--live', action='store_true')
    parser.add_argument('--share', action='store_true')
    parser.add_argument('--port', type=int)
    parser.add_argument('--disable-queue',
                        dest='enable_queue',
                        action='store_false')
    parser.add_argument('--allow-flagging', type=str, default='never')
    parser.add_argument('--allow-screenshot', action='store_true')
    return parser.parse_args()


def load_sample_image_paths() -> list[pathlib.Path]:
    image_dir = pathlib.Path('images')
    if not image_dir.exists():
        dataset_repo = 'hysts/sample-images-TADNE'
        path = huggingface_hub.hf_hub_download(dataset_repo,
                                               'images.tar.gz',
                                               repo_type='dataset',
                                               use_auth_token=TOKEN)
        with tarfile.open(path) as f:
            f.extractall()
    return sorted(image_dir.glob('*'))


def load_model(device: torch.device) -> torch.nn.Module:
    path = huggingface_hub.hf_hub_download(MODEL_REPO,
                                           MODEL_FILENAME,
                                           use_auth_token=TOKEN)
    state_dict = torch.load(path)
    model = torchvision.models.resnet50(num_classes=1062)
    model.load_state_dict(state_dict)
    model.to(device)
    model.eval()
    return model


def load_labels() -> list[str]:
    label_path = huggingface_hub.hf_hub_download(MODEL_REPO,
                                                 LABEL_FILENAME,
                                                 use_auth_token=TOKEN)
    with open(label_path) as f:
        labels = [line.strip() for line in f.readlines()]
    return labels


@torch.inference_mode()
def predict(image: PIL.Image.Image, score_threshold: float,
            device: torch.device, model: torch.nn.Module,
            labels: list[str]) -> dict[str, float]:
    data = ImageWrapper(image).resize_square(256).alpha_bg(
        c='w').convert('RGB').tensor()
    data = data.to(device).unsqueeze(0)

    preds = model(data)[0]
    preds = torch.sigmoid(preds)
    preds = preds.cpu().numpy().astype(float)

    res = dict()
    for prob, label in zip(preds, labels):
        if prob < score_threshold:
            continue
        res[label] = prob
    return res


def main():
    gr.close_all()

    args = parse_args()
    device = torch.device(args.device)

    image_paths = load_sample_image_paths()
    examples = [[path.as_posix(), args.score_threshold]
                for path in image_paths]

    model = load_model(device)
    labels = load_labels()

    func = functools.partial(predict,
                             device=device,
                             model=model,
                             labels=labels)
    func = functools.update_wrapper(func, predict)

    repo_url = 'https://github.com/ShuhongChen/bizarre-pose-estimator'
    title = 'ShuhongChen/bizarre-pose-estimator (tagger)'
    description = f'A demo for {repo_url}'
    article = None

    gr.Interface(
        func,
        [
            gr.inputs.Image(type='pil', label='Input'),
            gr.inputs.Slider(0,
                             1,
                             step=args.score_slider_step,
                             default=args.score_threshold,
                             label='Score Threshold'),
        ],
        gr.outputs.Label(label='Output'),
        theme=args.theme,
        title=title,
        description=description,
        article=article,
        examples=examples,
        allow_screenshot=args.allow_screenshot,
        allow_flagging=args.allow_flagging,
        live=args.live,
    ).launch(
        enable_queue=args.enable_queue,
        server_port=args.port,
        share=args.share,
    )


if __name__ == '__main__':
    main()