-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
261 lines (217 loc) · 8.29 KB
/
utils.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
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
import numpy as np
import matplotlib.cm as cm
import matplotlib.pyplot as plt
import matplotlib
import torch
K = 100
alpha = 0.8
def one_each(pred, thresh=0.0):
# Postprocess frcnn: get at most one instance per class
# Return: boxes and labels
conf = pred['scores'] > thresh
conf_scores = pred['scores'][conf]
conf_boxes = pred['boxes'][conf].int()
conf_labels = pred['labels'][conf].int()
valid = torch.zeros_like(conf_labels).bool()
unique_labels = torch.unique(conf_labels)
for uni in unique_labels:
p = (conf_labels==uni).nonzero(as_tuple=False).reshape(-1)
valid[p[0]] = True
pd_scores = conf_scores[valid]
pd_boxes = conf_boxes[valid]
pd_labels = conf_labels[valid]
return pd_boxes, pd_labels
def clean_heatmap(heatmap,mode=1):
'''
Normalize raw heatmap such that
- the entries are all nonnegative
- the entries sum up to 1.0
'''
if mode == 1:
min_val = np.min(heatmap)
heatmap = heatmap - min_val # make sure heatmap is always positive
med_val = np.median(heatmap) # take median
heatmap[heatmap < med_val] = 0 # get rid of all values below median
elif mode == 2:
min_val = np.min(heatmap)
if min_val < 0:
heatmap = heatmap - min_val
else:
raise RuntimeError('Unknown mode for cleaning heatmap')
heatmap = heatmap / np.sum(heatmap)
return heatmap
def topk_points(heatmap,k):
'''
Return the top k most likely keypoint detections in the heatmap
xy: xy coordinates of the keypoints
vk: values of the top k probabilities (re-normalized to sum up to 1.0)
'''
r, c = np.unravel_index(
np.flip(np.argsort(heatmap.ravel())), heatmap.shape)
v = heatmap[r,c]
rk = r[:k]
ck = c[:k]
vk = v[:k]
vk = vk / np.sum(vk)
# offset the coordinates to the center
# For example (0,0) pixel has coordinates (0.5,0.5)
ck = ck + 0.5
rk = rk + 0.5
xy = np.stack((ck,rk),axis=1)
return xy, vk
def conformity_score(kpt,heatmap,type="ball"):
'''
Given a keypoint location on a 2D image, and
a heatmap prediction of the keypoint location,
compute the nonconformility score
:param
kpt: (2,) numpy array
heatmap: (H,W) numpy array
type: choice of the conformity function
:return
conformity score
'''
heatmap = clean_heatmap(heatmap,mode=1)
if type == "ball":
r, c = np.unravel_index(
np.argmax(heatmap.ravel()),heatmap.shape)
maxp = heatmap[r,c]
# note here kpt loc (x,y), x corresponds to column, y corresponds to row!!!
r += 0.5
c += 0.5
dist = np.linalg.norm( kpt - np.array([c,r]) )
return dist * maxp
elif type == "ellipse":
xy, v = topk_points(heatmap,K)
wkpt = v @ xy
diff = xy - wkpt
sigma = diff.T @ np.diag(v) @ diff
sigmainv = np.linalg.inv(sigma)
return (kpt - wkpt) @ sigmainv @ (kpt-wkpt)
else:
raise RuntimeError('Unknown score type.')
def icp(heatmap,q,type="ball"):
'''
Given a heatmap and a quantile, output the inductive prediction set
:param
heatmap: numpy array H x W
q: scalar quantitle
type: choice of conformity function
'''
heatmap = clean_heatmap(heatmap,mode=1)
if type == "ball":
r, c = np.unravel_index(
np.argmax(heatmap.ravel()),heatmap.shape)
maxp = heatmap[r,c]
c += 0.5
r += 0.5
return np.array([c,r]), q / maxp # return center and radius
elif type == "ellipse":
xy, v = topk_points(heatmap,K)
wkpt = v @ xy
diff = xy - wkpt
sigma = diff.T @ np.diag(v) @ diff
sigmainv = np.linalg.inv(sigma)
return wkpt, sigmainv / q # return center and information matrix
else:
raise RuntimeError('Unknown score type.')
def draw_icp_ball(img,heatmaps,kpt_gt,pred_set,fname=None,show=False,heatmaponly=False):
linewidth = 2
pointsize = 2
height = 20
subplot_gap = 0.05
num_kpts = len(pred_set)
colors = cm.Set2(np.linspace(0, 1, num_kpts))
fig, axes = plt.subplots(1,num_kpts+1,figsize=(2*height,2*height))
fig.subplots_adjust(wspace=subplot_gap)
for i in range(num_kpts):
heatmap = np.squeeze(heatmaps[i,:,:])
heatmap = clean_heatmap(heatmap)
axes[i].imshow(img)
axes[i].imshow(heatmap,alpha=alpha)
if not heatmaponly:
center, radius = pred_set[i]
circ = plt.Circle(center,radius,color=colors[i],fill=True,linewidth=linewidth,alpha=0.5)
axes[i].add_patch(circ)
circ_b = plt.Circle(center,radius,color=colors[i],fill=False,linewidth=linewidth)
axes[i].add_patch(circ_b)
# point = plt.Circle((kpt_gt[i,0],kpt_gt[i,1]),pointsize,color=colors[i])
point = plt.Rectangle([kpt_gt[i,0]-pointsize/2,kpt_gt[i,1]-pointsize/2],pointsize,pointsize,color=colors[i])
axes[i].add_patch(point)
axes[i].xaxis.set_visible(False)
axes[i].yaxis.set_visible(False)
axes[-1].imshow(img)
for i in range(num_kpts):
center, radius = pred_set[i]
circ = plt.Circle(center,radius,color=colors[i],fill=True,linewidth=linewidth,alpha=0.5)
axes[-1].add_patch(circ)
circ_b = plt.Circle(center,radius,color=colors[i],fill=False,linewidth=linewidth)
axes[-1].add_patch(circ_b)
# point = plt.Circle((kpt_gt[i,0],kpt_gt[i,1]),pointsize,color=colors[i])
point = plt.Rectangle([kpt_gt[i,0]-pointsize/2,kpt_gt[i,1]-pointsize/2],pointsize,pointsize,color=colors[i])
axes[-1].add_patch(point)
axes[-1].xaxis.set_visible(False)
axes[-1].yaxis.set_visible(False)
if fname is not None:
plt.savefig(fname,bbox_inches='tight')
if show:
plt.show()
return fig
def angle_length_ellipse(A):
'''
Given an ellipse x' * A * x <= 1
return a, b, and angle
angle is the angle rotating from x to y (anti-clockwise)
'''
v, V = np.linalg.eig(A)
idx = np.argsort(v)
v = v[idx] # ascending order v[0] <= ... <= v[-1]
V = V[:,idx]
ab = np.sqrt(1.0 / v)
a = ab[0]
b = ab[-1]
assert a >= b, "semi-axes lengths wrong."
Vl = V[:,0] # long axis direction
angle = np.arctan2(Vl[-1],Vl[0]) / np.pi * 180.0
return a, b, angle
def draw_icp_ellipse(img,heatmaps,kpt_gt,pred_set,fname=None,show=False):
linewidth = 2
pointsize = 2
height = 20
subplot_gap = 0.05
num_kpts = len(pred_set)
colors = cm.Set2(np.linspace(0, 1, num_kpts))
fig, axes = plt.subplots(1,num_kpts+1,figsize=(2*height,2*height))
fig.subplots_adjust(wspace=subplot_gap)
for i in range(num_kpts):
heatmap = np.squeeze(heatmaps[i,:,:])
heatmap = clean_heatmap(heatmap)
axes[i].imshow(img)
axes[i].imshow(heatmap,alpha=alpha)
center, lam = pred_set[i]
a, b, angle = angle_length_ellipse(lam)
ellipse = matplotlib.patches.Ellipse(center,2*a,2*b,angle=angle,color=colors[i],fill=True,linewidth=linewidth,alpha=0.5)
axes[i].add_patch(ellipse)
ellipse_b = matplotlib.patches.Ellipse(center,2*a,2*b,angle=angle,color=colors[i],fill=False,linewidth=linewidth)
axes[i].add_patch(ellipse_b)
point = plt.Rectangle([kpt_gt[i,0]-pointsize/2,kpt_gt[i,1]-pointsize/2],pointsize,pointsize,color=colors[i])
axes[i].add_patch(point)
axes[i].xaxis.set_visible(False)
axes[i].yaxis.set_visible(False)
axes[-1].imshow(img)
for i in range(num_kpts):
center, lam = pred_set[i]
a, b, angle = angle_length_ellipse(lam)
ellipse = matplotlib.patches.Ellipse(center,2*a,2*b,angle=angle,color=colors[i],fill=True,linewidth=linewidth,alpha=0.5)
axes[-1].add_patch(ellipse)
ellipse_b = matplotlib.patches.Ellipse(center,2*a,2*b,angle=angle,color=colors[i],fill=False,linewidth=linewidth)
axes[-1].add_patch(ellipse_b)
point = plt.Rectangle([kpt_gt[i,0]-pointsize/2,kpt_gt[i,1]-pointsize/2],pointsize,pointsize,color=colors[i])
axes[-1].add_patch(point)
axes[-1].xaxis.set_visible(False)
axes[-1].yaxis.set_visible(False)
if fname is not None:
plt.savefig(fname,bbox_inches='tight')
if show:
plt.show()
return fig