-
Notifications
You must be signed in to change notification settings - Fork 2
/
EigenTheda.py
397 lines (305 loc) · 16.8 KB
/
EigenTheda.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
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
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
import copy
import numpy as np
from copy import deepcopy
from math import cos, sin, pi, atan2
import torch
# x y w h theda → EigenVector、EigenValue → Matrix: 2 × 3
# x y w h: 0~1, theda: 0~180
def xywhTheda2Eigen(input): # [:, xywhtheda] [a1, a2, cx; a3, a4, cy]
input[..., 4] = input[..., 4] / 180 * pi
# 2D
if len(input.shape) == 2:
output = np.zeros((input.shape[0], 2, 3))
output_vector = np.zeros((input.shape[0], 6)) # [cx, cy, a1, a2, a3, a4]
output_vector_5 = np.zeros((input.shape[0], 5)) # [cx, cy, z1, z3, z2]
EigenMatrix = np.zeros((input.shape[0], 2, 3))
EigenMatrix[:, 0, 2] = input[:, 0] # cx
EigenMatrix[:, 1, 2] = input[:, 1] # cy
EigenValueMatrix = np.zeros((input.shape[0], 2, 2))
EigenValueMatrix[:, 0, 0] = input[:, 2] # w
EigenValueMatrix[:, 1, 1] = input[:, 3] # h
theda = input[:, 4]
EigenVectorMatrix = np.zeros((input.shape[0], 2, 2))
EigenVectorMatrix[:, 0, 0] = np.cos(theda)
EigenVectorMatrix[:, 1, 0] = -np.sin(theda)
EigenVectorMatrix[:, 0, 1] = -np.sin(theda)
EigenVectorMatrix[:, 1, 1] = -np.cos(theda)
# EigenMatrix[:2, :2] = np.dot(np.dot(EigenVectorMatrix, EigenValueMatrix), EigenVectorMatrix.T)
E_sin_temp = np.einsum('ijk,ikr->ijr', EigenVectorMatrix, EigenValueMatrix)
EigenMatrix[:, :2, :2] = np.einsum('ijk,ikr->ijr', E_sin_temp, EigenVectorMatrix.transpose(0, 2, 1))
output[:, :, :] = EigenMatrix[:, :, :]
output_vector[:, 0] = output[:, 0, 2] # cx
output_vector[:, 1] = output[:, 1, 2] # cy
output_vector[:, 2] = output[:, 0, 0] # a1
output_vector[:, 3] = output[:, 0, 1] # a2
output_vector[:, 4] = output[:, 1, 0] # a3
output_vector[:, 5] = output[:, 1, 1] # a4
output_vector_5[:, 0] = output[:, 0, 2] # cx
output_vector_5[:, 1] = output[:, 1, 2] # cy
output_vector_5[:, 2] = output[:, 0, 0] # z1 0~1
output_vector_5[:, 3] = output[:, 1, 1] # z3 0~1
output_vector_5[:, 4] = output[:, 0, 1] # z2 -1~1
elif len(input.shape) == 3:
output_vector_5_3 = np.zeros((input.shape[0], input.shape[1], 5))
for index in range(input.shape[0]):
output = np.zeros((input.shape[1], 2, 3))
output_vector = np.zeros((input.shape[1], 6)) # [cx, cy, a1, a2, a3, a4]
output_vector_5 = np.zeros((input.shape[1], 5)) # [cx, cy, z1, z3, z2]
EigenMatrix = np.zeros((input.shape[1], 2, 3))
EigenMatrix[:, 0, 2] = input[index, :, 0] # cx
EigenMatrix[:, 1, 2] = input[index, :, 1] # cy
EigenValueMatrix = np.zeros((input.shape[1], 2, 2))
EigenValueMatrix[:, 0, 0] = input[index, :, 2] # w
EigenValueMatrix[:, 1, 1] = input[index, :, 3] # h
theda = input[index, :, 4]
EigenVectorMatrix = np.zeros((input.shape[1], 2, 2))
EigenVectorMatrix[:, 0, 0] = np.cos(theda)
EigenVectorMatrix[:, 1, 0] = -np.sin(theda)
EigenVectorMatrix[:, 0, 1] = -np.sin(theda)
EigenVectorMatrix[:, 1, 1] = -np.cos(theda)
# EigenMatrix[:2, :2] = np.dot(np.dot(EigenVectorMatrix, EigenValueMatrix), EigenVectorMatrix.T)
E_sin_temp = np.einsum('ijk,ikr->ijr', EigenVectorMatrix, EigenValueMatrix)
EigenMatrix[:, :2, :2] = np.einsum('ijk,ikr->ijr', E_sin_temp, EigenVectorMatrix.transpose(0, 2, 1))
output[:, :, :] = EigenMatrix[:, :, :]
output_vector[:, 0] = output[:, 0, 2] # cx
output_vector[:, 1] = output[:, 1, 2] # cy
output_vector[:, 2] = output[:, 0, 0] # a1
output_vector[:, 3] = output[:, 0, 1] # a2
output_vector[:, 4] = output[:, 1, 0] # a3
output_vector[:, 5] = output[:, 1, 1] # a4
output_vector_5[:, 0] = output[:, 0, 2] # cx
output_vector_5[:, 1] = output[:, 1, 2] # cy
output_vector_5[:, 2] = output[:, 0, 0] # z1 0~1
output_vector_5[:, 3] = output[:, 1, 1] # z3 0~1
output_vector_5[:, 4] = output[:, 0, 1] # z2 -1~1
output_vector_5_3[index, :, :] = output_vector_5[:, :]
output_vector_5 = output_vector_5_3
return output_vector_5
def xywhTheda2Eigen_numpy(input): # x y w h: 0~1, theda: 0~180 to x, y, z1, z3, z2
cx = copy.deepcopy(input[..., 0])
cy = copy.deepcopy(input[..., 1])
w = copy.deepcopy(input[..., 2])
h = copy.deepcopy(input[..., 3])
angle = copy.deepcopy(input[..., 4])
angle = angle / 180 * pi
sin_ = np.sin(angle)
cos_ = np.cos(angle)
z1 = w * (cos_**2) + h * (sin_**2)
z3 = w * (sin_**2) + h * (cos_**2)
z2 = (h - w) * sin_ * cos_
output = np.concatenate([cx.reshape(-1, 1), cy.reshape(-1, 1), z1.reshape(-1, 1), z3.reshape(-1, 1), z2.reshape(-1, 1)], axis=1)
return output
# EigenVector、EigenValue → Matrix: 2 × 3 →x y w h theda
# x y w h theda: 0~180
def Eigen2xywhTheda(input_vector): # [:, xywhtheda] [a1, a2, cx; a3, a4, cy]
# 2D
if len(input_vector.shape) == 2:
output = np.zeros((input_vector.shape[0], 5))
input = np.zeros((input_vector.shape[0], 2, 3)) # Vector to Matrix
input[:, 0, 2] = input_vector[:, 0] # cx
input[:, 1, 2] = input_vector[:, 1] # cy
input[:, 0, 0] = input_vector[:, 2] # a1
input[:, 0, 1] = input_vector[:, 4] # a2
input[:, 1, 0] = input_vector[:, 4] # a3
input[:, 1, 1] = input_vector[:, 3] # a4
output[:, 0] = input[:, 0, 2] # cx
output[:, 1] = input[:, 1, 2] # cy
values, vectors = np.linalg.eig(input[:, :, :2])
a = values.argmax(axis=1)
output[:, 2] = values[np.arange(a.size), a]
b = 1 - a
output[:, 3] = values[np.arange(b.size), b]
theda_vector = vectors[np.arange(a.size), :, a]
theda = np.arctan2(theda_vector[:, 1], theda_vector[:, 0]) + pi # 0 ~ 2*pi
theda[theda >= pi] -= pi
theda[theda == pi] -= pi # [0, 2*pi] to [0, pi)
output_theda = pi - theda
output_theda[output_theda == pi] -= pi # (0, pi] to [0, pi)
output[:, 4] = output_theda
elif len(input_vector.shape) == 3:
output_3 = np.zeros((input_vector.shape[0], input_vector.shape[1], 5))
for index in range(input_vector.shape[0]):
output = np.zeros((input_vector.shape[1], 5))
input = np.zeros((input_vector.shape[1], 2, 3)) # Vector to Matrix
input[:, 0, 2] = input_vector[index, :, 0] # cx
input[:, 1, 2] = input_vector[index, :, 1] # cy
input[:, 0, 0] = input_vector[index, :, 2] # a1
input[:, 0, 1] = input_vector[index, :, 4] # a2
input[:, 1, 0] = input_vector[index, :, 4] # a3
input[:, 1, 1] = input_vector[index, :, 3] # a4
output[:, 0] = input[:, 0, 2] # cx
output[:, 1] = input[:, 1, 2] # cy
values, vectors = np.linalg.eig(input[:, :, :2]) # np.where
a = values.argmax(axis=1)
output[:, 2] = values[np.arange(a.size), a]
b = 1 - a
output[:, 3] = values[np.arange(b.size), b]
theda_vector = vectors[np.arange(a.size), :, a]
theda = np.arctan2(theda_vector[:, 1], theda_vector[:, 0]) + pi # 0 ~ 2*pi
theda[theda >= pi] -= pi
theda[theda == pi] -= pi # [0, 2*pi] to [0, pi)
output_theda = pi - theda
output_theda[output_theda == pi] -= pi # (0, pi] to [0, pi)
output[:, 4] = output_theda
output_3[index, :, :] = output
output = output_3
output[..., 4] = output[..., 4] / pi * 180
return output
# x y w h theda: 0~180
def Eigen2xywhTheda_numpy(input_vector, h_thred=0, angle_thred=0.95): # input_vector: (n, 2) (cx, cy, z1, z3, z2) to (cx, cy, w, h, theda)
cx = copy.deepcopy(input_vector[..., 0])
cy = copy.deepcopy(input_vector[..., 1])
z1 = copy.deepcopy(input_vector[..., 2])
z2 = copy.deepcopy(input_vector[..., 4])
z3 = copy.deepcopy(input_vector[..., 3])
w = 0.5 * (z1 + z3 + ((z1 - z3) ** 2 + 4 * (z2 ** 2)) ** 0.5)
h = 0.5 * (z1 + z3 - ((z1 - z3) ** 2 + 4 * (z2 ** 2)) ** 0.5)
'h >= 0 means positive definiteness'
non_positive_define = (h <= h_thred)
sin_2theta = -2 * z2 / (w - h)
cos_2theta = (z1 - z3) / (w - h)
sin_2theta = np.clip(sin_2theta, -1, 1)
cos_2theta = np.clip(cos_2theta, -1, 1)
arcsin_2_theta = np.arcsin(sin_2theta) # [-pi/2, pi/2]
arccos_2_theta = np.arccos(cos_2theta) # [0, pi]
_2_theta = np.zeros_like(arcsin_2_theta)
arcsin_2_theta_1 = (arcsin_2_theta >= 0) * (arcsin_2_theta <= pi / 2) * (arccos_2_theta >= 0) * (arccos_2_theta <= pi / 2)
_2_theta[arcsin_2_theta_1] = arcsin_2_theta[arcsin_2_theta_1]
arcsin_2_theta_2 = (arcsin_2_theta >= 0) * (arcsin_2_theta <= pi / 2) * (arccos_2_theta > pi / 2) * (arccos_2_theta <= pi)
_2_theta[arcsin_2_theta_2] = pi - arcsin_2_theta[arcsin_2_theta_2]
arcsin_2_theta_3 = (arcsin_2_theta >= -pi / 2) * (arcsin_2_theta < 0) * (arccos_2_theta > pi / 2) * (arccos_2_theta <= pi)
_2_theta[arcsin_2_theta_3] = pi - arcsin_2_theta[arcsin_2_theta_3]
arcsin_2_theta_4 = (arcsin_2_theta >= -pi / 2) * (arcsin_2_theta < 0) * (arccos_2_theta >= 0) * (arccos_2_theta <= pi / 2)
_2_theta[arcsin_2_theta_4] = 2 * pi + arcsin_2_theta[arcsin_2_theta_4]
theta = _2_theta / 2 # [0, pi)
theta = theta / pi * 180 # [0, 180)
theta = np.clip(theta, 1, 179)
cx, cy, w, h, theta = cx.reshape(-1, 1), cy.reshape(-1, 1), w.reshape(-1, 1), h.reshape(-1, 1), theta.reshape(-1, 1)
output = np.concatenate([cx, cy, w, h, theta], axis=1)
return output, non_positive_define
def arcsin_taylor(xi, x0):
x = torch.arcsin(x0) + ((1 - x0**2)**(-0.5)) * (xi - x0) + x0 * ((1 - x0**2)**(-1.5)) * (xi - x0)**2
return x
def arccos_taylor(xi, x0):
x = torch.arccos(x0) - ((1 - x0**2)**(-0.5)) * (xi - x0) - x0 * ((1 - x0**2)**(-1.5)) * (xi - x0)**2
return x
# x y w h theda: 0~180
def Eigen2xywhTheda_numpy_tensor(input_vector, h_thred=0, angle_thred=1.0): # input_vector: (n, 2) (cx, cy, z1, z3, z2) to (cx, cy, w, h, theda)
cx = input_vector[..., 0]
cy = input_vector[..., 1]
z1 = input_vector[..., 2]
z2 = input_vector[..., 4]
z3 = input_vector[..., 3]
w = 0.5 * (z1 + z3 + ((z1 - z3) ** 2 + 4 * (z2 ** 2)) ** 0.5)
h = 0.5 * (z1 + z3 - ((z1 - z3) ** 2 + 4 * (z2 ** 2)) ** 0.5)
'h >= 0 means positive definiteness'
non_positive_define = (h <= h_thred)
w[(w / h < 1.05) & (h > 0)] *= 1.05
sin_2theta = -2 * z2 / (w - h)
cos_2theta = (z1 - z3) / (w - h)
non_positive_define |= (sin_2theta > angle_thred) | (sin_2theta < -angle_thred) | (cos_2theta > angle_thred) | (cos_2theta < -angle_thred)
sin_2theta = torch.clip(sin_2theta, -angle_thred, angle_thred)
cos_2theta = torch.clip(cos_2theta, -angle_thred, angle_thred)
arcsin_2_theta = torch.arcsin(sin_2theta) # [-pi/2, pi/2]
arccos_2_theta = torch.arccos(cos_2theta) # [0, pi]
_2_theta = torch.zeros_like(arcsin_2_theta)
arcsin_2_theta_1 = (arcsin_2_theta >= 0) * (arcsin_2_theta <= pi / 2) * (arccos_2_theta >= 0) * (arccos_2_theta <= pi / 2)
_2_theta[arcsin_2_theta_1] = arcsin_2_theta[arcsin_2_theta_1]
arcsin_2_theta_2 = (arcsin_2_theta >= 0) * (arcsin_2_theta <= pi / 2) * (arccos_2_theta > pi / 2) * (arccos_2_theta <= pi)
_2_theta[arcsin_2_theta_2] = pi - arcsin_2_theta[arcsin_2_theta_2]
arcsin_2_theta_3 = (arcsin_2_theta >= -pi / 2) * (arcsin_2_theta < 0) * (arccos_2_theta > pi / 2) * (arccos_2_theta <= pi)
_2_theta[arcsin_2_theta_3] = pi - arcsin_2_theta[arcsin_2_theta_3]
arcsin_2_theta_4 = (arcsin_2_theta >= -pi / 2) * (arcsin_2_theta < 0) * (arccos_2_theta >= 0) * (arccos_2_theta <= pi / 2)
_2_theta[arcsin_2_theta_4] = 2 * pi + arcsin_2_theta[arcsin_2_theta_4]
theta = _2_theta / 2 # [0, pi)
theta = theta / pi * 180 # [0, 180)
theta = torch.clip(theta, 1, 179)
# cx, cy, w, h, theta = cx.view(-1, 1), cy.view(-1, 1), w.view(-1, 1), h.view(-1, 1), theta.view(-1, 1)
cx, cy, w, h, theta = cx.unsqueeze(-1), cy.unsqueeze(-1), w.unsqueeze(-1), h.unsqueeze(-1), theta.unsqueeze(-1)
output = torch.cat([cx, cy, w, h, theta], dim=-1)
# output = output.to(dtype=original_type)
return output, non_positive_define
def Eigen2xywhTheda_gpu(input_vector): # [:, xywhtheda] [a1, a2, cx; a3, a4, cy]
# 2D
if len(input_vector.shape) == 2:
output = torch.zeros((input_vector.shape[0], 5))
input = torch.zeros((input_vector.shape[0], 2, 3)) # Vector to Matrix
input[:, 0, 2] = input_vector[:, 0] # cx
input[:, 1, 2] = input_vector[:, 1] # cy
input[:, 0, 0] = input_vector[:, 2] # a1
input[:, 0, 1] = input_vector[:, 4] # a2
input[:, 1, 0] = input_vector[:, 4] # a3
input[:, 1, 1] = input_vector[:, 3] # a4
output[:, 0] = input[:, 0, 2] # cx
output[:, 1] = input[:, 1, 2] # cy
values, vectors = torch.linalg.eigh(input[:, :, :2]) # np.where
a = values.argmax(axis=1)
output[:, 2] = values[torch.arange(len(a)), a]
b = 1 - a
output[:, 3] = values[torch.arange(len(b)), b]
theda_vector = vectors[torch.arange(len(a)), :, a]
theda = torch.arctan2(theda_vector[:, 1], theda_vector[:, 0]) + pi # 0 ~ 2*pi
theda[theda >= pi] -= pi
theda[theda == pi] -= pi # [0, 2*pi] to [0, pi)
output_theda = pi - theda
output_theda[output_theda == pi] -= pi # (0, pi] to [0, pi)
output[:, 4] = output_theda
elif len(input_vector.shape) == 3:
output_3 = torch.zeros((input_vector.shape[0], input_vector.shape[1], 5))
for index in range(input_vector.shape[0]):
output = torch.zeros((input_vector.shape[1], 5))
input = torch.zeros((input_vector.shape[1], 2, 3)) # Vector to Matrix
input[:, 0, 2] = input_vector[index, :, 0] # cx
input[:, 1, 2] = input_vector[index, :, 1] # cy
input[:, 0, 0] = input_vector[index, :, 2] # a1
input[:, 0, 1] = input_vector[index, :, 4] # a2
input[:, 1, 0] = input_vector[index, :, 4] # a3
input[:, 1, 1] = input_vector[index, :, 3] # a4
output[:, 0] = input[:, 0, 2] # cx
output[:, 1] = input[:, 1, 2] # cy
values, vectors = torch.linalg.eigh(input[:, :, :2]) # np.where
a = values.argmax(axis=1)
output[:, 2] = values[torch.arange(a.size), a]
b = 1 - a
output[:, 3] = values[torch.arange(b.size), b]
theda_vector = vectors[torch.arange(a.size), :, a]
theda = torch.arctan2(theda_vector[:, 1], theda_vector[:, 0]) + pi # 0 ~ 2*pi
theda[theda >= pi] -= pi
theda[theda == pi] -= pi # [0, 2*pi] to [0, pi)
output_theda = pi - theda
output_theda[output_theda == pi] -= pi # (0, pi] to [0, pi)
output[:, 4] = output_theda
output_3[index, :, :] = output
output = output_3
output[..., 4] = output[..., 4] / pi * 180
return output
def Eigen2x(input): # clsid, cx, cy, z1, z3, z2 to clsid, cx, cy, x1, x2, x3
output = deepcopy(input)
output[..., -3] = (input[..., -3] + input[..., -2]) / 2
output[..., -2] = input[..., -3]
output[..., -1] = (input[..., -3] + input[..., -2] + input[..., -1] * 2) / 2
return output
def x2Eigen(input): # clsid, cx, cy, x1, x2, x3 to clsid, cx, cy, z1, z3, z2
if torch.is_tensor(input):
output = torch.zeros_like(input).to(input.device)
else:
output = np.zeros_like(input)
output[..., :-3] = input[..., :-3]
output[..., -3] = input[..., -2]
output[..., -2] = input[..., -3] * 2 - input[..., -2]
output[..., -1] = input[..., -1] - input[..., -3]
return output
if __name__ == '__main__':
input = np.array([[[0.3, 0.4, 0.8, 0.5, 179 ],
[0.1, 0.2, 0.41, 0.4, 0 ],
[0.3, 0.4, 0.7, 0.6, 90]],
[[0.3, 0.4, 0.8, 0.5, 134],
[0.1, 0.2, 0.41, 0.4, 50],
[0.3, 0.4, 0.7, 0.6, 70]]]
)
output = xywhTheda2Eigen(input)
print("input: ", input.shape, input)
print("output: ", output.shape, output)
output2 = Eigen2xywhTheda(output)
print("output: ", output.shape, output)
print("output2: ", output2.shape, output2)