File size: 13,729 Bytes
5ac1897
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
from lib.kits.basic import *

# Copied from: https://github.com/shubham-goel/4D-Humans/blob/6ec79656a23c33237c724742ca2a0ec00b398b53/hmr2/datasets/utils.py#L663-L944

def crop_to_hips(center_x: float, center_y: float, width: float, height: float, keypoints_2d: np.ndarray) -> Tuple:
    """
    Extreme cropping: Crop the box up to the hip locations.
    Args:
        center_x (float): x coordinate of the bounding box center.
        center_y (float): y coordinate of the bounding box center.
        width (float): Bounding box width.
        height (float): Bounding box height.
        keypoints_2d (np.ndarray): Array of shape (N, 3) containing 2D keypoint locations.
    Returns:
        center_x (float): x coordinate of the new bounding box center.
        center_y (float): y coordinate of the new bounding box center.
        width (float): New bounding box width.
        height (float): New bounding box height.
    """
    keypoints_2d = keypoints_2d.copy()
    lower_body_keypoints = [10, 11, 13, 14, 19, 20, 21, 22, 23, 24, 25+0, 25+1, 25+4, 25+5]
    keypoints_2d[lower_body_keypoints, :] = 0
    if keypoints_2d[:, -1].sum() > 1:
        center, scale = get_bbox(keypoints_2d)
        center_x = center[0]
        center_y = center[1]
        width = 1.1 * scale[0]
        height = 1.1 * scale[1]
    return center_x, center_y, width, height


def crop_to_shoulders(center_x: float, center_y: float, width: float, height: float, keypoints_2d: np.ndarray):
    """
    Extreme cropping: Crop the box up to the shoulder locations.
    Args:
        center_x (float): x coordinate of the bounding box center.
        center_y (float): y coordinate of the bounding box center.
        width (float): Bounding box width.
        height (float): Bounding box height.
        keypoints_2d (np.ndarray): Array of shape (N, 3) containing 2D keypoint locations.
    Returns:
        center_x (float): x coordinate of the new bounding box center.
        center_y (float): y coordinate of the new bounding box center.
        width (float): New bounding box width.
        height (float): New bounding box height.
    """
    keypoints_2d = keypoints_2d.copy()
    lower_body_keypoints = [3, 4, 6, 7, 8, 9, 10, 11, 12, 13, 14, 19, 20, 21, 22, 23, 24] + [25 + i for i in [0, 1, 2, 3, 4, 5, 6, 7, 10, 11, 14, 15, 16]]
    keypoints_2d[lower_body_keypoints, :] = 0
    center, scale = get_bbox(keypoints_2d)
    if keypoints_2d[:, -1].sum() > 1:
        center, scale = get_bbox(keypoints_2d)
        center_x = center[0]
        center_y = center[1]
        width = 1.2 * scale[0]
        height = 1.2 * scale[1]
    return center_x, center_y, width, height


def crop_to_head(center_x: float, center_y: float, width: float, height: float, keypoints_2d: np.ndarray):
    """
    Extreme cropping: Crop the box and keep on only the head.
    Args:
        center_x (float): x coordinate of the bounding box center.
        center_y (float): y coordinate of the bounding box center.
        width (float): Bounding box width.
        height (float): Bounding box height.
        keypoints_2d (np.ndarray): Array of shape (N, 3) containing 2D keypoint locations.
    Returns:
        center_x (float): x coordinate of the new bounding box center.
        center_y (float): y coordinate of the new bounding box center.
        width (float): New bounding box width.
        height (float): New bounding box height.
    """
    keypoints_2d = keypoints_2d.copy()
    lower_body_keypoints = [3, 4, 6, 7, 8, 9, 10, 11, 12, 13, 14, 19, 20, 21, 22, 23, 24] + [25 + i for i in [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 14, 15, 16]]
    keypoints_2d[lower_body_keypoints, :] = 0
    if keypoints_2d[:, -1].sum() > 1:
        center, scale = get_bbox(keypoints_2d)
        center_x = center[0]
        center_y = center[1]
        width = 1.3 * scale[0]
        height = 1.3 * scale[1]
    return center_x, center_y, width, height


def crop_torso_only(center_x: float, center_y: float, width: float, height: float, keypoints_2d: np.ndarray):
    """
    Extreme cropping: Crop the box and keep on only the torso.
    Args:
        center_x (float): x coordinate of the bounding box center.
        center_y (float): y coordinate of the bounding box center.
        width (float): Bounding box width.
        height (float): Bounding box height.
        keypoints_2d (np.ndarray): Array of shape (N, 3) containing 2D keypoint locations.
    Returns:
        center_x (float): x coordinate of the new bounding box center.
        center_y (float): y coordinate of the new bounding box center.
        width (float): New bounding box width.
        height (float): New bounding box height.
    """
    keypoints_2d = keypoints_2d.copy()
    nontorso_body_keypoints = [0, 3, 4, 6, 7, 10, 11, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24] + [25 + i for i in [0, 1, 4, 5, 6, 7, 10, 11, 13, 17, 18]]
    keypoints_2d[nontorso_body_keypoints, :] = 0
    if keypoints_2d[:, -1].sum() > 1:
        center, scale = get_bbox(keypoints_2d)
        center_x = center[0]
        center_y = center[1]
        width = 1.1 * scale[0]
        height = 1.1 * scale[1]
    return center_x, center_y, width, height


def crop_rightarm_only(center_x: float, center_y: float, width: float, height: float, keypoints_2d: np.ndarray):
    """
    Extreme cropping: Crop the box and keep on only the right arm.
    Args:
        center_x (float): x coordinate of the bounding box center.
        center_y (float): y coordinate of the bounding box center.
        width (float): Bounding box width.
        height (float): Bounding box height.
        keypoints_2d (np.ndarray): Array of shape (N, 3) containing 2D keypoint locations.
    Returns:
        center_x (float): x coordinate of the new bounding box center.
        center_y (float): y coordinate of the new bounding box center.
        width (float): New bounding box width.
        height (float): New bounding box height.
    """
    keypoints_2d = keypoints_2d.copy()
    nonrightarm_body_keypoints = [0, 1, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24] + [25 + i for i in [0, 1, 2, 3, 4, 5, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18]]
    keypoints_2d[nonrightarm_body_keypoints, :] = 0
    if keypoints_2d[:, -1].sum() > 1:
        center, scale = get_bbox(keypoints_2d)
        center_x = center[0]
        center_y = center[1]
        width = 1.1 * scale[0]
        height = 1.1 * scale[1]
    return center_x, center_y, width, height


def crop_leftarm_only(center_x: float, center_y: float, width: float, height: float, keypoints_2d: np.ndarray):
    """
    Extreme cropping: Crop the box and keep on only the left arm.
    Args:
        center_x (float): x coordinate of the bounding box center.
        center_y (float): y coordinate of the bounding box center.
        width (float): Bounding box width.
        height (float): Bounding box height.
        keypoints_2d (np.ndarray): Array of shape (N, 3) containing 2D keypoint locations.
    Returns:
        center_x (float): x coordinate of the new bounding box center.
        center_y (float): y coordinate of the new bounding box center.
        width (float): New bounding box width.
        height (float): New bounding box height.
    """
    keypoints_2d = keypoints_2d.copy()
    nonleftarm_body_keypoints = [0, 1, 2, 3, 4, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24] + [25 + i for i in [0, 1, 2, 3, 4, 5, 6, 7, 8, 12, 13, 14, 15, 16, 17, 18]]
    keypoints_2d[nonleftarm_body_keypoints, :] = 0
    if keypoints_2d[:, -1].sum() > 1:
        center, scale = get_bbox(keypoints_2d)
        center_x = center[0]
        center_y = center[1]
        width = 1.1 * scale[0]
        height = 1.1 * scale[1]
    return center_x, center_y, width, height


def crop_legs_only(center_x: float, center_y: float, width: float, height: float, keypoints_2d: np.ndarray):
    """
    Extreme cropping: Crop the box and keep on only the legs.
    Args:
        center_x (float): x coordinate of the bounding box center.
        center_y (float): y coordinate of the bounding box center.
        width (float): Bounding box width.
        height (float): Bounding box height.
        keypoints_2d (np.ndarray): Array of shape (N, 3) containing 2D keypoint locations.
    Returns:
        center_x (float): x coordinate of the new bounding box center.
        center_y (float): y coordinate of the new bounding box center.
        width (float): New bounding box width.
        height (float): New bounding box height.
    """
    keypoints_2d = keypoints_2d.copy()
    nonlegs_body_keypoints = [0, 1, 2, 3, 4, 5, 6, 7, 15, 16, 17, 18] + [25 + i for i in [6, 7, 8, 9, 10, 11, 12, 13, 15, 16, 17, 18]]
    keypoints_2d[nonlegs_body_keypoints, :] = 0
    if keypoints_2d[:, -1].sum() > 1:
        center, scale = get_bbox(keypoints_2d)
        center_x = center[0]
        center_y = center[1]
        width = 1.1 * scale[0]
        height = 1.1 * scale[1]
    return center_x, center_y, width, height


def crop_rightleg_only(center_x: float, center_y: float, width: float, height: float, keypoints_2d: np.ndarray):
    """
    Extreme cropping: Crop the box and keep on only the right leg.
    Args:
        center_x (float): x coordinate of the bounding box center.
        center_y (float): y coordinate of the bounding box center.
        width (float): Bounding box width.
        height (float): Bounding box height.
        keypoints_2d (np.ndarray): Array of shape (N, 3) containing 2D keypoint locations.
    Returns:
        center_x (float): x coordinate of the new bounding box center.
        center_y (float): y coordinate of the new bounding box center.
        width (float): New bounding box width.
        height (float): New bounding box height.
    """
    keypoints_2d = keypoints_2d.copy()
    nonrightleg_body_keypoints = [0, 1, 2, 3, 4, 5, 6, 7, 8, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21] + [25 + i for i in [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18]]
    keypoints_2d[nonrightleg_body_keypoints, :] = 0
    if keypoints_2d[:, -1].sum() > 1:
        center, scale = get_bbox(keypoints_2d)
        center_x = center[0]
        center_y = center[1]
        width = 1.1 * scale[0]
        height = 1.1 * scale[1]
    return center_x, center_y, width, height

def crop_leftleg_only(center_x: float, center_y: float, width: float, height: float, keypoints_2d: np.ndarray):
    """
    Extreme cropping: Crop the box and keep on only the left leg.
    Args:
        center_x (float): x coordinate of the bounding box center.
        center_y (float): y coordinate of the bounding box center.
        width (float): Bounding box width.
        height (float): Bounding box height.
        keypoints_2d (np.ndarray): Array of shape (N, 3) containing 2D keypoint locations.
    Returns:
        center_x (float): x coordinate of the new bounding box center.
        center_y (float): y coordinate of the new bounding box center.
        width (float): New bounding box width.
        height (float): New bounding box height.
    """
    keypoints_2d = keypoints_2d.copy()
    nonleftleg_body_keypoints = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 15, 16, 17, 18, 22, 23, 24] + [25 + i for i in [0, 1, 2, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18]]
    keypoints_2d[nonleftleg_body_keypoints, :] = 0
    if keypoints_2d[:, -1].sum() > 1:
        center, scale = get_bbox(keypoints_2d)
        center_x = center[0]
        center_y = center[1]
        width = 1.1 * scale[0]
        height = 1.1 * scale[1]
    return center_x, center_y, width, height


def full_body(keypoints_2d: np.ndarray) -> bool:
    """
    Check if all main body joints are visible.
    Args:
        keypoints_2d (np.ndarray): Array of shape (N, 3) containing 2D keypoint locations.
    Returns:
        bool: True if all main body joints are visible.
    """

    body_keypoints_openpose = [2, 3, 4, 5, 6, 7, 10, 11, 13, 14]
    body_keypoints = [25 + i for i in [8, 7, 6, 9, 10, 11, 1, 0, 4, 5]]
    return (np.maximum(keypoints_2d[body_keypoints, -1], keypoints_2d[body_keypoints_openpose, -1]) > 0).sum() == len(body_keypoints)


def upper_body(keypoints_2d: np.ndarray):
    """
    Check if all upper body joints are visible.
    Args:
        keypoints_2d (np.ndarray): Array of shape (N, 3) containing 2D keypoint locations.
    Returns:
        bool: True if all main body joints are visible.
    """
    lower_body_keypoints_openpose = [10, 11, 13, 14]
    lower_body_keypoints = [25 + i for i in [1, 0, 4, 5]]
    upper_body_keypoints_openpose = [0, 1, 15, 16, 17, 18]
    upper_body_keypoints = [25+8, 25+9, 25+12, 25+13, 25+17, 25+18]
    return ((keypoints_2d[lower_body_keypoints + lower_body_keypoints_openpose, -1] > 0).sum() == 0)\
       and ((keypoints_2d[upper_body_keypoints + upper_body_keypoints_openpose, -1] > 0).sum() >= 2)


def get_bbox(keypoints_2d: np.ndarray, rescale: float = 1.2) -> Tuple:
    """
    Get center and scale for bounding box from openpose detections.
    Args:
        keypoints_2d (np.ndarray): Array of shape (N, 3) containing 2D keypoint locations.
        rescale (float): Scale factor to rescale bounding boxes computed from the keypoints.
    Returns:
        center (np.ndarray): Array of shape (2,) containing the new bounding box center.
        scale (float): New bounding box scale.
    """
    valid = keypoints_2d[:,-1] > 0
    valid_keypoints = keypoints_2d[valid][:,:-1]
    center = 0.5 * (valid_keypoints.max(axis=0) + valid_keypoints.min(axis=0))
    bbox_size = (valid_keypoints.max(axis=0) - valid_keypoints.min(axis=0))
    # adjust bounding box tightness
    scale = bbox_size
    scale *= rescale
    return center, scale