File size: 7,615 Bytes
37c98fe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
179
180
181
182
183
184
185
186
187
188
189
190
from __future__ import annotations

import os
import subprocess
import sys

if os.getenv('SYSTEM') == 'spaces':
    import mim

    mim.uninstall('mmcv-full', confirm_yes=True)
    mim.install('mmcv-full==1.5.0', is_yes=True)

    subprocess.call('pip uninstall -y opencv-python'.split())
    subprocess.call('pip uninstall -y opencv-python-headless'.split())
    subprocess.call('pip install opencv-python-headless==4.5.5.64'.split())

import huggingface_hub
import numpy as np
import torch
import torch.nn as nn

sys.path.insert(0, 'ViTPose/')

from mmdet.apis import inference_detector, init_detector
from mmpose.apis import (inference_top_down_pose_model, init_pose_model,
                         process_mmdet_results, vis_pose_result)

HF_TOKEN = os.environ['HF_TOKEN']


class DetModel:
    def __init__(self, device: str | torch.device):
        self.device = torch.device(device)
        self.models = self._load_models()
        self.model_name = 'YOLOX-l'

    def _load_models(self) -> dict[str, nn.Module]:
        model_dict = {
            'YOLOX-tiny': {
                'config':
                'mmdet_configs/configs/yolox/yolox_tiny_8x8_300e_coco.py',
                'model':
                'https://download.openmmlab.com/mmdetection/v2.0/yolox/yolox_tiny_8x8_300e_coco/yolox_tiny_8x8_300e_coco_20211124_171234-b4047906.pth',
            },
            'YOLOX-s': {
                'config':
                'mmdet_configs/configs/yolox/yolox_s_8x8_300e_coco.py',
                'model':
                'https://download.openmmlab.com/mmdetection/v2.0/yolox/yolox_s_8x8_300e_coco/yolox_s_8x8_300e_coco_20211121_095711-4592a793.pth',
            },
            'YOLOX-l': {
                'config':
                'mmdet_configs/configs/yolox/yolox_l_8x8_300e_coco.py',
                'model':
                'https://download.openmmlab.com/mmdetection/v2.0/yolox/yolox_l_8x8_300e_coco/yolox_l_8x8_300e_coco_20211126_140236-d3bd2b23.pth',
            },
            'YOLOX-x': {
                'config':
                'mmdet_configs/configs/yolox/yolox_x_8x8_300e_coco.py',
                'model':
                'https://download.openmmlab.com/mmdetection/v2.0/yolox/yolox_x_8x8_300e_coco/yolox_x_8x8_300e_coco_20211126_140254-1ef88d67.pth',
            },
        }
        models = {
            key: init_detector(dic['config'], dic['model'], device=self.device)
            for key, dic in model_dict.items()
        }
        return models

    def set_model_name(self, name: str) -> None:
        self.model_name = name

    def detect_and_visualize(
            self, image: np.ndarray,
            score_threshold: float) -> tuple[list[np.ndarray], np.ndarray]:
        out = self.detect(image)
        vis = self.visualize_detection_results(image, out, score_threshold)
        return out, vis

    def detect(self, image: np.ndarray) -> list[np.ndarray]:
        image = image[:, :, ::-1]  # RGB -> BGR
        model = self.models[self.model_name]
        out = inference_detector(model, image)
        return out

    def visualize_detection_results(
            self,
            image: np.ndarray,
            detection_results: list[np.ndarray],
            score_threshold: float = 0.3) -> np.ndarray:
        person_det = [detection_results[0]] + [np.array([]).reshape(0, 5)]

        image = image[:, :, ::-1]  # RGB -> BGR
        model = self.models[self.model_name]
        vis = model.show_result(image,
                                person_det,
                                score_thr=score_threshold,
                                bbox_color=None,
                                text_color=(200, 200, 200),
                                mask_color=None)
        return vis[:, :, ::-1]  # BGR -> RGB


class PoseModel:
    def __init__(self, device: str | torch.device):
        self.device = torch.device(device)
        self.models = self._load_models()
        self.model_name = 'ViTPose-B (multi-task train, COCO)'

    def _load_models(self) -> dict[str, nn.Module]:
        model_dict = {
            'ViTPose-B (single-task train)': {
                'config':
                'ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/ViTPose_base_coco_256x192.py',
                'model': 'models/vitpose-b.pth',
            },
            'ViTPose-L (single-task train)': {
                'config':
                'ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/ViTPose_large_coco_256x192.py',
                'model': 'models/vitpose-l.pth',
            },
            'ViTPose-B (multi-task train, COCO)': {
                'config':
                'ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/ViTPose_base_coco_256x192.py',
                'model': 'models/vitpose-b-multi-coco.pth',
            },
            'ViTPose-L (multi-task train, COCO)': {
                'config':
                'ViTPose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/ViTPose_large_coco_256x192.py',
                'model': 'models/vitpose-l-multi-coco.pth',
            },
        }
        models = dict()
        for key, dic in model_dict.items():
            ckpt_path = huggingface_hub.hf_hub_download(
                'hysts/ViTPose', dic['model'], use_auth_token=HF_TOKEN)
            model = init_pose_model(dic['config'],
                                    ckpt_path,
                                    device=self.device)
            models[key] = model
        return models

    def set_model_name(self, name: str) -> None:
        self.model_name = name

    def predict_pose_and_visualize(
        self,
        image: np.ndarray,
        det_results: list[np.ndarray],
        box_score_threshold: float,
        kpt_score_threshold: float,
        vis_dot_radius: int,
        vis_line_thickness: int,
    ) -> tuple[list[dict[str, np.ndarray]], np.ndarray]:
        out = self.predict_pose(image, det_results, box_score_threshold)
        vis = self.visualize_pose_results(image, out, kpt_score_threshold,
                                          vis_dot_radius, vis_line_thickness)
        return out, vis

    def predict_pose(
            self,
            image: np.ndarray,
            det_results: list[np.ndarray],
            box_score_threshold: float = 0.5) -> list[dict[str, np.ndarray]]:
        image = image[:, :, ::-1]  # RGB -> BGR
        model = self.models[self.model_name]
        person_results = process_mmdet_results(det_results, 1)
        out, _ = inference_top_down_pose_model(model,
                                               image,
                                               person_results=person_results,
                                               bbox_thr=box_score_threshold,
                                               format='xyxy')
        return out

    def visualize_pose_results(self,
                               image: np.ndarray,
                               pose_results: list[np.ndarray],
                               kpt_score_threshold: float = 0.3,
                               vis_dot_radius: int = 4,
                               vis_line_thickness: int = 1) -> np.ndarray:
        image = image[:, :, ::-1]  # RGB -> BGR
        model = self.models[self.model_name]
        vis = vis_pose_result(model,
                              image,
                              pose_results,
                              kpt_score_thr=kpt_score_threshold,
                              radius=vis_dot_radius,
                              thickness=vis_line_thickness)
        return vis[:, :, ::-1]  # BGR -> RGB