-
Notifications
You must be signed in to change notification settings - Fork 3
Expand file tree
/
Copy pathtoolkit.py
More file actions
207 lines (173 loc) · 6.47 KB
/
toolkit.py
File metadata and controls
207 lines (173 loc) · 6.47 KB
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
import numpy as np
import object_detection2.bboxes as odb
import math
from data_types import TrackState
def kps_dis(kp0,kp1,bbox0,bbox1,min_nr=4,inf_dis=1e8):
score0 = kp0[:,-1]
score1 = kp1[:,-1]
mask0 = score0>0.5
mask1 = score1>0.5
base_mask = np.ones([17],dtype=np.bool)
base_mask[[7,8,9,10,13,14,15,16]] = False
mask = np.logical_and(mask0,mask1)
#mask = np.logical_and(mask,base_mask)
total_nr = np.count_nonzero(mask)
wh0 = bbox0[2:]-bbox0[:2]
wh1 = bbox1[2:]-bbox1[:2]
S = np.min((wh0+wh1)/2)
if total_nr < min_nr:
dis = 1.0-odb.npbboxes_jaccard([bbox0],[bbox1])
return dis[0]
delta = (kp1[mask]-kp0[mask])[:,:2]
delta = np.square(delta)
delta = np.sum(delta,axis=-1)
delta = np.sqrt(delta)
mean_dis = np.mean(delta)
dis = mean_dis/S
dis = min(max(dis,0),1.0)
return dis
def kps_dis_matrix(kps0,kps1,bboxes0,bboxes1):
kps0 = np.array(kps0)
kps1 = np.array(kps1)
dis_matrix = np.ones([kps0.shape[0],kps1.shape[0]],dtype=np.float32)
for i in range(kps0.shape[0]):
for j in range(kps1.shape[0]):
dis_matrix[i,j] = kps_dis(kps0[i],kps1[j],bboxes0[i],bboxes1[j])
return dis_matrix
def kps_bboxes_nms(bboxes,kps,threshold=0.1,iou_threshold=0.2):
if len(bboxes)<2:
return bboxes,kps
kps_matrix = kps_dis_matrix(kps,kps,bboxes,bboxes)
nr_bboxes = len(bboxes)
mask = np.ones([nr_bboxes],dtype=np.bool)
for i in range(nr_bboxes-1):
ious = odb.npbboxes_jaccard([bboxes[i]],bboxes)
for j in range(i+1,nr_bboxes):
if ious[j]<iou_threshold:
continue
if kps_matrix[i,j]<threshold:
mask[j] = False
mask = mask.tolist()
return bboxes[mask],kps[mask],mask
def log_print(*args):
#print(*args)
pass
def remove_half_kps(bboxes,kps):
if len(bboxes)<=1:
return [True]
kps = kps[:,[13,14,15,16],:]
kps_score = kps[...,-1]
keep = np.sum(np.array(kps_score>0.3).astype(np.int32),axis=-1)>1
keep_bboxes = bboxes[keep.tolist()]
for i in range(len(bboxes)):
if keep[i]:
continue
bbox = bboxes[i]
gious_dis = (1-odb.npgiou([bbox],keep_bboxes))/2
if np.any(gious_dis<0.65):
keep[i] = True
return keep
def align_kps(kp0,kp1,threshold=0.4):
base_mask = np.ones([17],dtype=np.bool)
base_mask[[7,8,9,10,13,14,15,16]] = False
mask0 = kp0[...,-1]>threshold
mask1 = kp1[...,-1]>threshold
mask = np.logical_and(np.logical_and(mask0,mask1),base_mask)
if np.any(mask):
idx = np.argmax(mask)
else:
mask = np.logical_and(mask0,mask1)
if not np.any(mask):
return None,None
idx = np.argmax(mask)
kp0[...,:2] = kp0[...,:2]-np.expand_dims(kp0[idx,:2],axis=0)
kp1[...,:2] = kp1[...,:2]-np.expand_dims(kp1[idx,:2],axis=0)
return kp0,kp1
def simple_kps_dis(kp0,kp1,threshold=0.4):
mask0 = kp0[...,-1]>threshold
mask1 = kp1[...,-1]>threshold
mask = np.logical_and(mask0,mask1)
mask = mask.tolist()
if not np.any(mask):
return -1
dis = (kp0[mask]-kp1[mask])[...,:2]
dis = np.square(dis)
dis = np.sum(dis,axis=-1)
dis = np.mean(dis)
return math.sqrt(dis)
def simple_aligned_kps_dis(kp0,kp1,threshold=0.4):
kp0,kp1 = align_kps(kp0,kp1,threshold=threshold)
if kp0 is None:
return -1
return simple_kps_dis(kp0,kp1,threshold=threshold)
def kps_distance(atracks, btracks):
"""
Compute cost based on IoU
:type atracks: list[STrack]
:type btracks: list[STrack]
:rtype cost_matrix np.ndarray
"""
bbox0 = [track.ltrb for track in atracks]
bbox1 = [track.ltrb for track in btracks]
kps0 = [track.cur_kps for track in atracks]
kps1 = [track.cur_kps for track in btracks]
return kps_dis_matrix(kps0,kps1,bbox0,bbox1)
def set_untracked_dis(dis_matrix,tracks,v=1.0):
if dis_matrix.shape[0] != len(tracks):
print(f"ERROR size {dis_matrix.shape[0]} vs {len(tracks)}")
return dis_matrix
res = dis_matrix.copy()
for i,track in enumerate(tracks):
if(track.state == TrackState.Lost):
res[i] = v
return res
def gather_tracks_by_idx(tracks_pool,idxs):
'''
idx in idxs may not in stracks_pool
'''
res = []
for track in tracks_pool:
if track.track_idx in idxs:
res.append(track)
return res
if __name__ == "__main__":
kps0 = np.array([[ 487.1, 464.06, 0.45359],
[ 487.1, 458.19, 0.53741],
[ 483.19, 462.1, 0.46846],
[ 487.1, 458.19, 0.61229],
[ 475.35, 462.1, 0.60248],
[ 491.02, 454.27, 0.54884],
[ 473.4, 475.81, 0.3269],
[ 494.94, 436.65, 0.49745],
[ 469.48, 487.56, 0.31506],
[ 504.73, 420.98, 0.43875],
[ 475.35, 516.94, 0.48977],
[ 518.44, 513.02, 0.5471],
[ 502.77, 511.06, 0.63261],
[ 555.65, 514.98, 0.63434],
[ 520.4, 497.35, 0.80151],
[ 586.98, 524.77, 0.66719],
[ 549.77, 520.85, 0.44264]], dtype=np.float32)
kps1 = np.array([[ 497.93, 417.93, 0.35145],
[ 495.71, 414.6, 0.38472],
[ 495.71, 416.82, 0.29749],
[ 496.82, 411.27, 0.54249],
[ 502.37, 404.62, 0.25975],
[ 501.26, 409.05, 0.59756],
[ 522.34, 400.18, 0.4783],
[ 512.35, 433.46, 0.43057],
[ 547.85, 404.62, 0.51891],
[ 510.13, 462.3, 0.37314],
[ 562.27, 410.16, 0.47261],
[ 530.1, 427.91, 0.62482],
[ 548.96, 423.48, 0.65236],
[ 521.23, 457.87, 0.71844],
[ 537.87, 457.87, 0.8253],
[ 523.45, 497.8, 0.70386],
[ 542.3, 493.37, 0.64326]], dtype=np.float32)
bbox0 = np.array([ 461.47, 412.7, 617.43, 568.98], dtype=np.float32)
bbox1 = np.array([ 491.34, 385.63, 557.83, 503.51], dtype=np.float32)
print(odb.npbboxes_jaccard([bbox0],[bbox1]))
dis = kps_dis_matrix(np.array([kps0]),np.array([kps1]),[bbox0],[bbox1])
print(dis)
#print(simple_aligned_kps_dis(kps0,kps1))