forked from scnuhealthy/Tensorflow_PersonLab
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdata_pred.py
122 lines (99 loc) · 5.22 KB
/
data_pred.py
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
import numpy as np
import cv2
from config import config
map_shape = (config.IMAGE_SHAPE[0], config.IMAGE_SHAPE[1])
idx = np.rollaxis(np.indices(map_shape[::-1]), 0, 3).transpose((1,0,2))
def map_coco_to_personlab(keypoints):
permute = [0, 6, 8, 10, 5, 7, 9, 12, 14, 16, 11, 13, 15, 2, 1, 4, 3]
if len(keypoints.shape) == 2:
return keypoints[permute, :]
return keypoints[:, permute, :]
def get_ground_truth(instance_masks, all_keypoints):
assert(instance_masks.shape[-1] == len(all_keypoints))
discs = get_keypoint_discs(all_keypoints)
kp_maps = make_keypoint_maps(all_keypoints, discs)
short_offsets = compute_short_offsets(all_keypoints, discs)
mid_offsets = compute_mid_offsets(all_keypoints, discs)
long_offsets = compute_long_offsets(all_keypoints, instance_masks)
return kp_maps, short_offsets, mid_offsets, long_offsets
# return (人数,key数,401,401) 绘制heatmap的响应圆
def get_keypoint_discs(all_keypoints):
map_shape = (config.IMAGE_SHAPE[0], config.IMAGE_SHAPE[1])
discs = [[] for _ in range(len(all_keypoints))]
for i in range(config.NUM_KP):
centers = [keypoints[i,:2] for keypoints in all_keypoints if keypoints[i,2] > 0]
dists = np.zeros(map_shape+(len(centers),))
for k, center in enumerate(centers):
dists[:,:,k] = np.sqrt(np.square(center-idx).sum(axis=-1))
if len(centers) > 0:
inst_id = dists.argmin(axis=-1)
count = 0
for j in range(len(all_keypoints)):
if all_keypoints[j][i,2] > 0:
discs[j].append(np.logical_and(inst_id==count, dists[:,:,count]<=config.KP_RADIUS))
count +=1
else:
discs[j].append(np.array([]))
return discs
def make_keypoint_maps(all_keypoints, discs):
# map_shape = (config.IMAGE_SHAPE[0], config.IMAGE_SHAPE[1])
kp_maps = np.zeros(map_shape+(config.NUM_KP,))
for i in range(config.NUM_KP):
for j in range(len(discs)):
if all_keypoints[j][i,2] > 0:
kp_maps[discs[j][i], i] = 1.
return kp_maps
def compute_short_offsets(all_keypoints, discs):
# map_shape = (config.IMAGE_SHAPE[0], config.IMAGE_SHAPE[1])
r = config.KP_RADIUS
x = np.tile(np.arange(r, -r-1, -1), [2*r+1, 1])
y = x.transpose()
m = np.sqrt(x*x + y*y) <= r
kp_circle = np.stack([x, y], axis=-1) * np.expand_dims(m, axis=-1)
def copy_with_border_check(map, center, disc):
from_top = max(r-center[1], 0)
from_left = max(r-center[0], 0)
from_bottom = max(r-(map_shape[0]-center[1])+1, 0)
from_right = max(r-(map_shape[1]-center[0])+1, 0)
cropped_disc = disc[center[1]-r+from_top:center[1]+r+1-from_bottom, center[0]-r+from_left:center[0]+r+1-from_right]
map[center[1]-r+from_top:center[1]+r+1-from_bottom, center[0]-r+from_left:center[0]+r+1-from_right, :][cropped_disc,:] = \
kp_circle[from_top:2*r+1-from_bottom, from_left:2*r+1-from_right, :][cropped_disc,:]
offsets = np.zeros(map_shape+(2*config.NUM_KP,))
for i in range(config.NUM_KP):
# this_offset = np.zeros(shape=map_shape+(2,))
for j in range(len(all_keypoints)):
if all_keypoints[j][i,2] > 0:
copy_with_border_check(offsets[:,:,2*i:2*i+2], (all_keypoints[j][i,0], all_keypoints[j][i,1]), discs[j][i])
return offsets
def compute_mid_offsets(all_keypoints, discs):
# map_shape = (config.IMAGE_SHAPE[0], config.IMAGE_SHAPE[1])
offsets = np.zeros(map_shape+(4*config.NUM_EDGES,))
for i, edge in enumerate((config.EDGES + [edge[::-1] for edge in config.EDGES])):
#this_offset = np.zeros(map_shape+(2,))
for j in range(len(all_keypoints)):
if all_keypoints[j][edge[0],2] > 0 and all_keypoints[j][edge[1],2] > 0:
# idx = np.rollaxis(np.indices(map_shape), 0, 3).transpose((1,0,2))
# dists = np.array([[ all_keypoints[j][edge[1],0], all_keypoints[j][edge[1],1] ]]) - idx
m = discs[j][edge[0]]
dists = [[ all_keypoints[j][edge[1],0], all_keypoints[j][edge[1],1] ]] - idx[m,:]
# this_offset[m,:] = dists[m,:]
# offsets[:,:,2*i:2*i+2] = this_offset
offsets[m,2*i:2*i+2] = dists
return offsets
def compute_long_offsets(all_keypoints, instance_masks):
# map_shape = (config.IMAGE_SHAPE[0], config.IMAGE_SHAPE[1])
instance_masks = instance_masks.astype('bool')
offsets = np.zeros(map_shape+(2*config.NUM_KP,))
for i in range(config.NUM_KP):
# this_offset = np.zeros(map_shape+(2,))
for j in range(len(all_keypoints)):
if all_keypoints[j][i,2] > 0:
# idx = np.rollaxis(np.indices(map_shape), 0, 3).transpose((1,0,2))
m = instance_masks[:,:,j]
dists = [[ all_keypoints[j][i,0], all_keypoints[j][i,1] ]] - idx[m,:]
#this_offset[m,:] = dists[m,:]
offsets[m, 2*i:2*i+2] = dists
# offsets[:,:,2*i:2*i+2]
overlap = np.sum(instance_masks, axis=-1) >= 2
offsets[overlap,:] = 0.
return offsets