-
Notifications
You must be signed in to change notification settings - Fork 4
/
BPConvCopy.py
560 lines (474 loc) · 26.2 KB
/
BPConvCopy.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
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
# CopyRight by BUAA, VR-Lab
# Author : Chen Lang
# 2018/08/02
from numpy import *
import numpy as np
# from numba import jit
from get_descriptor import *
import matplotlib.pyplot as plt
from numba import jit
posLow = [(-2, -2), (-2, -1), (-2, 0), (-2, 1), (-2, 2), \
(-1, -2), (-1, -1), (-1, 0), (-1, 1), (-1, 2), \
(0, -2), (0, -1), (0, 0), (0, 1), (0, 2), \
(1, -2), (1, -1), (1, 0), (1, 1), (1, 2), \
(2, -2), (2, -1), (2, 0), (2, 1), (2, 2)]
posHigh = [(-1, -1), (-1, 0), (-1, 1), \
(0, -1), (0, 0), (0, 1), \
(1, -1), (1, 0), (1, 1)]
color = ['b', 'c', 'g', 'k', 'm', 'r', 'w', 'y']
eps = 0.0001
beta = 0.35 # 0.35
threshold = 0.55 # 0.4,0.7
def color_map(i):
colors = [
[255,0,0],
[0,255,0],
[0,0,255],
[128,128,0],
[0,128,128]
]
if i < 5:
return colors[i]
else:
return np.random.randint(0, 256, 3)
def extractPatch(coor, featureMap):
# coor:list size is K * 4, stands for the leftUP and rightBottom Corner of every patch to search
# featureMap: array size is row * col * Channel
patch = list()
for item in coor:
patch.append(featureMap[item[0]:item[2]+1, item[1]:item[3]+1, :])
return patch
def appendImage(imLeft, imRight):
# select the image with the fewest rows and fill in enough empty rows
# rowsNew = max(imLeft.shape[0], imRight.shape[0])
# colNew = max(imLeft.shape[1], imRight.shape[1])
rows1 = imLeft.shape[0]
rows2 = imRight.shape[0]
if len(imLeft.shape) == 3:
assert imLeft.shape[2] == imRight.shape[2]
depth = imLeft.shape[2]
if rows1 < rows2:
pad = ones((rows2-rows1, imLeft.shape[1], depth))*255
imLeft = concatenate([imLeft, pad], axis=0).astype(uint8) # 0为行扩展
elif rows1 > rows2:
pad = ones((rows1 - rows2, imRight.shape[1], depth))*255
imRight = concatenate([imRight, pad], axis=0).astype(uint8) # 数据类型转换要用arrayName.astype(dtype)
elif len(imLeft.shape) == 2:
if rows1 < rows2:
pad = ones((rows2 - rows1, imLeft.shape[1])) * 255
imLeft = concatenate([imLeft, pad], axis=0).astype(uint8) # 0为行扩展
elif rows1 > rows2:
pad = ones((rows1 - rows2, imRight.shape[1])) * 255
imRight = concatenate([imRight, pad], axis=0).astype(uint8) # 数据类型转换要用arrayName.astype(dtype)
imFinal = concatenate((imLeft, imRight), axis=1).astype(uint8) # 列扩展
# plt.imshow(imFinal)
# plt.axis('off')
# plt.show()
return imFinal
def pyramidPlot(upFilter, reluFiveLeft, reluFiveRight, step):
imgTotalSource = appendImage(reluFiveLeft[:, :, 0], reluFiveRight[:, :, 0])
plt.imshow(imgTotalSource)
flag = 0
for item in upFilter:
plt.plot(int(item[1]), int(item[0]), 'ro')
plt.plot(int(item[3]) + step+1, int(item[2]), 'ro')
plt.plot([int(item[1]), int(item[3]) + step+1], [int(item[0]), int(item[2])], color[flag % 8])
flag += 1
plt.colorbar()
plt.axis('off')
# plt.show()
def pyramidFiveRelu(reluFiveLeft, reluFiveRight):
# reluFiveLeft: type is numpy.array and the value is relu5_1, size is row * col * Channel
# reluFiveRight: type is numpy.array and the value is relu5_1, size is row1 * col1 * Channel
# return is the matching local area in pyramid Four
assert reluFiveLeft.shape[2] == reluFiveRight.shape[2]
#####################print the begin information#############
print("pyramidFiveRelu:\n")
print("reluFiveLeft shape is :{}\n".format(reluFiveLeft.shape))
print("reluFiveRight shape is :{}\n".format(reluFiveRight.shape))
#####################print the begin information#############
meanA, meanB = np.zeros((1, reluFiveLeft.shape[2])).squeeze(), np.zeros((1, reluFiveRight.shape[2])).squeeze()
stdA, stdB = np.zeros((1, reluFiveLeft.shape[2])).squeeze(), np.zeros((1, reluFiveRight.shape[2])).squeeze()
for i in range(reluFiveLeft.shape[2]):
meanA[i], stdA[i] = reluFiveLeft[:, :, i].mean() + eps, reluFiveLeft[:, :, i].std() + eps
meanB[i], stdB[i] = reluFiveRight[:, :, i].mean() + eps, reluFiveRight[:, :, i].std() + eps
meanPQ, stdPQ = (meanA + meanB) / 2, (stdA + stdB) / 2
# print(meanA, stdA)
'''
CaPQ = reluFiveLeft # ((reluFiveLeft - meanA) / stdA) * stdPQ + meanPQ # size is row * col * Channel
CbQP = reluFiveRight # ((reluFiveRight - meanB) / stdB) * stdPQ + meanPQ # with size (row, col, Channel)
'''
CaPQ = ((reluFiveLeft - meanA) / stdA) * stdPQ + meanPQ # size is row * col * Channel
CbQP = ((reluFiveRight - meanB) / stdB) * stdPQ + meanPQ # with size (row, col, Channel)
assert meanPQ.shape == meanA.shape and stdPQ.shape == meanB.shape # with size (1, Channel)
assert CaPQ.shape == reluFiveLeft.shape and CbQP.shape == reluFiveRight.shape
level = 1
pad = 0
# CaPQ, CbQP = np.pad(CaPQ, ((1,1),(1,1),(0,0)), 'constant'), np.pad(CbQP, ((1,1),(1,1),(0,0)), 'constant')
ANNTable = np.zeros((reluFiveLeft.shape[0], reluFiveLeft.shape[1])) # 存储A图区域中每个节点最近的B节点序号
L2A = np.zeros((reluFiveLeft.shape[0], reluFiveLeft.shape[1]))
L2B = np.zeros((reluFiveRight.shape[0], reluFiveRight.shape[1]))
for i in range(0, reluFiveLeft.shape[0], 1): # row
for j in range(0, reluFiveLeft.shape[1], 1): # col
L2A[i][j] = sqrt(sum(reluFiveLeft[i][j]**2))
L2B[i][j] = np.sqrt(sum(reluFiveRight[i][j] ** 2))
HA = (L2A - np.min(L2A))/(np.max(L2A)-np.min(L2A))
HB = (L2B - np.min(L2B))/(np.max(L2B)-np.min(L2B))
for i in range(0, reluFiveLeft.shape[0], 1): # row
for j in range(0, reluFiveLeft.shape[1], 1): # col
if HA[i][j] < beta:
ANNTable[i][j] = -1
continue
scoreTable = np.zeros((reluFiveRight.shape[0], reluFiveRight.shape[1])) # 为每个A图节点建立一个B图的打分表
for ii in range(0, reluFiveRight.shape[0], 1):
for jj in range(0, reluFiveRight.shape[1], 1):
for dPos in posHigh:
if (i+dPos[0])<0 or (ii+dPos[0])<0 or (i+dPos[0])>= reluFiveLeft.shape[0] or (j+dPos[1])>=reluFiveLeft.shape[1] \
or (j+dPos[1])<0 or (jj+dPos[1])<0 or (ii+dPos[0])>= reluFiveRight.shape[0] or (jj+dPos[1])>=reluFiveRight.shape[1]:
continue
else:
score = sum(CaPQ[i+dPos[0]][j+dPos[1]] * CbQP[ii+dPos[0]][jj+dPos[1]])\
/ (sqrt(sum(CbQP[ii+dPos[0]][jj+dPos[1]])**2)*sqrt(sum(CaPQ[i+dPos[0]][j+dPos[1]]**2)))
scoreTable[ii][jj] += score
ANNTable[i][j] = np.argmax(scoreTable) # A图当前节点在B中的搜索最佳位置
BNNTable = np.zeros((reluFiveRight.shape[0], reluFiveRight.shape[1])) # 存储B图区域中每个节点最近的A节点序号
for i in range(0, reluFiveRight.shape[0], 1): # row
for j in range(0, reluFiveRight.shape[1], 1): # col
if HB[i][j] < beta:
BNNTable[i][j] = -1
continue
scoreTable = np.zeros((reluFiveLeft.shape[0], reluFiveLeft.shape[1])) # 为每个B图节点建立一个A图的打分表
for ii in range(0, reluFiveLeft.shape[0],1):
for jj in range(0, reluFiveLeft.shape[1],1):
for dPos in posHigh:
if (i + dPos[0]) < 0 or (ii + dPos[0]) < 0 or (ii + dPos[0]) >= reluFiveLeft.shape[0] or (jj + dPos[1]) >= reluFiveLeft.shape[1] \
or (j + dPos[1]) < 0 or (jj + dPos[1]) < 0 or (i + dPos[0]) >= reluFiveRight.shape[0] or (j + dPos[1]) >= reluFiveRight.shape[1]:
continue
else:
score = sum(CbQP[i+dPos[0]][j+dPos[1]] * CaPQ[ii+dPos[0]][jj+dPos[1]])\
/ (np.sqrt(sum(CbQP[i+dPos[0]][j+dPos[1]])**2) * np.sqrt(sum(CaPQ[ii+dPos[0]][jj+dPos[1]]**2)))
scoreTable[ii][jj] += score
BNNTable[i][j] = np.argmax(scoreTable) # B图当前节点在A中搜索最佳位置
upFive = list()
for i in range(ANNTable.shape[0]):
for j in range(ANNTable.shape[1]):
if ANNTable[i][j] > -1:
rowB, colB = divmod(ANNTable[i][j], reluFiveRight.shape[1])
# print(rowB, colB)
rowA, colA = divmod(BNNTable[int(rowB)][int(colB)], reluFiveLeft.shape[1])
if rowA < 0 or colA < 0:
continue
rowA, colA = int(rowA), int(colA)
rowB, colB = int(rowB), int(colB)
if rowA == i and colA == j:
ans = [rowA, colA, rowB, colB] # find a corresponding region
upFive.append(ans)
else:
continue
##########################filter##############################
# print(HA.max(),HA.min(),HB.min(),HB.max())
# plt.figure()
# plt.subplot(1,2,1)
# plt.imshow(HA)
# plt.subplot(1,2,2)
# plt.imshow(HB)
# plt.show()
upFiveFilter = list()
for item in upFive:
if HA[item[0]][item[1]] > beta and HB[item[2]][item[3]] > beta and (item not in upFiveFilter):
upFiveFilter.append(item) # find a corresponding region which fits the threshold
else:
continue
##########################filter##############################
#####################print the end information################
print("pyramid " + str(5) + " end !\n")
print("The number of NBB matching points is : {}\n".format(len(upFiveFilter)))
print("The correspondence regions in Layer_4 is : {}\n".format(len(upFiveFilter)))
return upFiveFilter
#####################print the end information################
def draw_square(image, center, color, radius = 2):
d = 2*radius + 1
image_p = np.pad(image, ((radius,radius),(radius,radius),(0,0)),'constant')
center_p = [center[0]+radius, center[1]+radius]
image_p[center_p[0]-radius, (center_p[1]-radius):(center_p[1]-radius+d), :] = np.tile(color,[d,1])
image_p[(center_p[0]-radius):(center_p[0]-radius+d), center_p[1]-radius, :] = np.tile(color,[d,1])
image_p[center_p[0]+radius, (center_p[1]-radius):(center_p[1]-radius+d), :] = np.tile(color,[d,1])
image_p[(center_p[0]-radius):(center_p[0]-radius+d), center_p[1]+radius, :] = np.tile(color,[d,1])
return image_p[radius:image_p.shape[0]-radius, radius:image_p.shape[1]-radius, :]
def drawCorrespondence(imgA, imgB, upFiveFilter, radius):
# [rowA, colA, rowB, colB]
A_marked, B_marked = imgA.copy(), imgB.copy()
scale = 16
flag = 0
for i in range(len(upFiveFilter)):
color = color_map(i)
center_1 = [upFiveFilter[i][z] * scale for z in range(2)]
center_2 = [upFiveFilter[i][z] * scale for z in range(2, 4)]
A_marked = draw_square(A_marked, [center_1[0] + radius, center_1[1] + radius], color, radius=radius)
B_marked = draw_square(B_marked, [center_2[0] + radius, center_2[1] + radius], color, radius=radius)
flag += 1
plt.figure(), plt.imshow(A_marked)
plt.figure(), plt.imshow(B_marked)
plt.axis('off')
#################################
# imgTotal = appendImage(imgA, imgB)
batch = np.concatenate((imgA, imgB), 1)
color = ['b', 'c', 'g', 'k', 'm', 'r', 'w', 'y']
flag = 0
plt.figure(), plt.imshow(batch)
plt.axis('off')
for i in range(len(upFiveFilter)):
center_1 = [upFiveFilter[i][z] * scale for z in range(2)]
center_2 = [upFiveFilter[i][z] * scale for z in range(2, 4)]
plt.plot(center_1[1], center_1[0], color[flag % 8] + '.') # show points
plt.plot(224 + center_2[1], center_2[0], color[flag % 8] + '.') # show points
plt.plot([center_1[1], 224 + center_2[1]], [center_1[0], center_2[0]], color[flag % 8]) # True matching
flag += 1
#################################
plt.show()
def pyramidFive(reluFiveLeft, reluFiveRight):
upFiveFilter = pyramidFiveRelu(reluFiveLeft, reluFiveRight)
matchFourPatch = list()
# drawCorrespondence(imgA, imgB, upFiveFilter, 8)
# upFiveFilter = pyramidRansac(upFiveFilter, reluFiveLeft, reluFiveRight, 14, 1.0) # use ransac to filter out some points
for item in upFiveFilter:
#########################the Patch in A########################
LowApx, highApx = 2*item[0] - 3, 2*item[0] + 3
LowApy, highApy = 2*item[1] - 3, 2*item[1] + 3
LowBpx, highBpx = 2*item[2] - 3, 2*item[2] + 3
LowBpy, highBpy = 2*item[3] - 3, 2*item[3] + 3
matchFourPatch.append([LowApx, LowApy, highApx, highApy, LowBpx, LowBpy, highBpx, highBpy]) # lowRow,lowCol,highRow,highCol
#########################the Patch in A########################
return upFiveFilter, matchFourPatch
@jit
def pyramidRelu(item, convLeft, convRight, level, leftFlag, rightFlag):
# item = [LowApx, LowApy, highApx, highApy, LowBpx, LowBpy, highBpx, highBpy]
# convLeft: type is np.array with size (row, col, Channel), means the relu_1 of pyramid_level
# convRight: type is np.array with size (row1, col1, Channel), means the relu_1 of prramid_level
# print(item)
# print("{} layer is {}*{}".format(level, item[2]-item[0], item[3]-item[1]))
assert len(item) == 8
assert convLeft.shape[2] == convRight.shape[2]
assert (item[2]-item[0]) == (item[6]-item[4]) and (item[3]-item[1]) == (item[7]-item[5])
#####################print the begin information#############
# print("pyramid" + str(level) + "Relu:\n")
# print("reluFiveLeft shape is :{}\n".format(convLeft.shape))
# print("reluFiveRight shape is :{}\n".format(convRight.shape))
#####################print the begin information#############
meanA, stdA = np.zeros((1, convLeft.shape[2])).squeeze(), np.zeros((1, convLeft.shape[2])).squeeze()
meanB, stdB = np.zeros((1, convRight.shape[2])).squeeze(), np.zeros((1, convRight.shape[2])).squeeze()
for i in range(convLeft.shape[2]):
# print(convLeft[item[0]:(item[2]+1), item[1]:(item[3]+1), i])
meanA[i] = convLeft[item[0]:(item[2]+1), item[1]:(item[3]+1), i].mean() + eps
stdA[i] = convLeft[item[0]:1+item[2], item[1]:1+item[3], i].std() + eps
meanB[i] = convRight[item[4]:1+item[6], item[5]:1+item[7], i].mean() + eps
stdB[i] = convRight[item[4]:1+item[6], item[5]:1+item[7], i].std() + eps
# print(meanA)
meanPQ, stdPQ = (meanA + meanB) / 2, (stdA + stdB) / 2 # 5 * 5 * Channel or 7 * 7 * Channel
CaPQ = ((convLeft - meanA) / stdA) * stdPQ + meanPQ
CbQP = ((convRight - meanB) / stdB) * stdPQ + meanPQ
assert CaPQ.shape == convLeft.shape and CbQP.shape == convRight.shape
assert meanPQ.shape == stdPQ.shape and meanA.shape == meanB.shape
pos = posHigh if level >= 3 else posLow
# r = 1
# pad = 0 if level >= 4 else 2
# item = [LowApx, LowApy, highApx, highApy, LowBpx, LowBpy, highBpx, highBpy]
ANNTable = np.zeros((item[2]-item[0]+1, item[3]-item[1]+1))
L2A = np.zeros((item[2]-item[0]+1, item[3]-item[1]+1))
L2B = np.zeros((item[6] - item[4] + 1, item[7] - item[5] + 1))
# starttime = datetime.datetime.now()
for i in range(item[0], item[2] + 1, 1):
for j in range(item[1], item[3] + 1, 1):
L2A[i-item[0]][j-item[1]] = np.sqrt(sum(convLeft[i][j]**2))
L2B[i-item[0]][j-item[1]] = np.sqrt(sum(convRight[i-item[0]+item[4]][j-item[1]+item[5]] ** 2))
HaP = (L2A - L2A.min())/(L2A.max()-L2A.min())
HbQ = (L2B - L2B.min())/(L2B.max()-L2B.min())
for i in range(item[0], item[2]+1, 1):
for j in range(item[1], item[3]+1, 1):
# if leftFlag[i][j]:
# ANNTable[i - item[0]][j - item[1]] = -1
# continue
leftFlag[i][j] = 1
if HaP[i-item[0]][j-item[1]] < threshold:
ANNTable[i - item[0]][j - item[1]] = -1
continue
scoreTable = np.zeros((item[6]-item[4]+1, item[7]-item[5]+1))
for ii in range(item[4], item[6]+1, 1):
for jj in range(item[5], item[7]+1, 1):
for dPos in pos:
if (i + dPos[0]) < 0 or (ii + dPos[0]) < 0 or (i + dPos[0]) >= convLeft.shape[0] or (j + dPos[1]) >= convLeft.shape[1] \
or (j + dPos[1]) < 0 or (jj + dPos[1]) < 0 or (ii + dPos[0]) >= convRight.shape[0] or (jj + dPos[1]) >= convRight.shape[1]:
continue
else:
score = sum(CaPQ[i+dPos[0]][j+dPos[1]]*CbQP[ii+dPos[0]][jj+dPos[1]])/(\
np.sqrt(sum(CaPQ[i+dPos[0]][j+dPos[1]]**2)*sum(CbQP[ii+dPos[0]][jj+dPos[1]]**2)))
scoreTable[ii-item[4]][jj-item[5]] += score
ANNTable[i-item[0]][j-item[1]] = np.argmax(scoreTable) # 存储的是相对(item[0], item[1])的相对位置序号
# endtime1 = datetime.datetime.now()
# print("nns is :",endtime1-endtime)
BNNTable = np.zeros((item[6] - item[4] + 1, item[7] - item[5] + 1))
# L2B = np.zeros((item[6] - item[4] + 1, item[7] - item[5] + 1))
for i in range(item[4], item[6] + 1, 1):
for j in range(item[5], item[7] + 1, 1):
# if rightFlag[i][j]:
# BNNTable[i - item[4]][j - item[5]] = -1
# continue
rightFlag[i][j] = 1
if HbQ[i-item[4]][j-item[5]] < threshold:
BNNTable[i - item[4]][j - item[5]] = -1
continue
scoreTable = np.zeros((item[2] - item[0] + 1, item[3] - item[1] + 1))
for ii in range(item[0], item[2] + 1, 1):
for jj in range(item[1], item[3] + 1, 1):
for dPos in pos:
if (i + dPos[0]) < 0 or (ii + dPos[0]) < 0 or (ii + dPos[0]) >= convLeft.shape[0] or (jj + dPos[1]) >= convLeft.shape[1] \
or (j + dPos[1]) < 0 or (jj + dPos[1]) < 0 or (i + dPos[0]) >= convRight.shape[0] or (j + dPos[1]) >= convRight.shape[1]:
continue
else:
score = sum(CbQP[i + dPos[0]][j + dPos[1]] * CaPQ[ii + dPos[0]][jj + dPos[1]]) / ( \
np.sqrt(sum(CbQP[i+dPos[0]][j+dPos[1]]**2) * sum(CaPQ[ii+dPos[0]][jj+dPos[1]]**2)))
scoreTable[ii-item[0]][jj-item[1]] += score
BNNTable[i-item[4]][j-item[5]] = np.argmax(scoreTable) # 存储的是相对(item[4], item[5])的相对位置序号
# L2B[i - item[4]][j - item[5]] = np.sqrt(sum(convRight[i][j] ** 2))
upCoor = list()
for i in range(ANNTable.shape[0]):
for j in range(ANNTable.shape[1]):
if ANNTable[i][j] > -1:
rowB, colB = divmod(ANNTable[i][j], item[7]-item[5]+1)
rowA, colA = divmod(BNNTable[int(rowB)][int(colB)], item[3]-item[1]+1)
if rowA < 0 or colA < 0:
continue
rowA, colA = rowA + item[0], colA + item[1]
rowB, colB = rowB + item[4], colB + item[5]
rowA, colA = int(rowA), int(colA)
rowB, colB = int(rowB), int(colB)
if rowA == (i+item[0]) and colA == (j+item[1]):
upCoor.append([rowA, colA, rowB, colB])
else:
continue
##########################filter##############################
upFilter = list()
# meanH, meanL = (L2A.max() + L2B.max()) / 2, (L2A.min() + L2B.min()) / 2
# beta = meanL + 0.6 * (meanH - meanL)
# 0.5 used for source and 0.4 used for cropped image
'''
HaP = (L2A - L2A.min())/(L2A.max()-L2A.min())
HbQ = (L2B - L2B.min())/(L2B.max()-L2A.min())
# plt.figure()
# plt.subplot(1,2,1)
# plt.imshow(HaP)
# plt.subplot(1,2,2)
# plt.imshow(HbQ)
# plt.show()
for ele in upCoor:
if HaP[ele[0]-item[0]][ele[1]-item[1]] > threshold and HbQ[ele[2]-item[4]][ele[3]-item[5]] > threshold:
upFilter.append(ele)
else:
continue
##########################filter##############################
#####################print the end information################
'''
static = 8 if level >=3 else 5 if level >=2 else 2
if len(upCoor) > static:
ans = []
alist = list(np.random.randint(0, len(upCoor), static))
for item in alist:
ans.append(upCoor[item])
upCoor = ans
# print("cur return size is:", len(upCoor))
return upCoor, leftFlag, rightFlag
#####################print the end information################
@jit
def pyramidLevel(matchtPatchCoor, convLeft, convRight, level):
# matchtFourPatchCoor: list size is K * 8, K is the number of the patch to search in feature map of level four
# 8 means the flaged coordinates of matching patch to search in two feature map
# convLeft: array size is row * col * Channel, is the left feature map
# convRight:array with size row * col * Channel, is the right feature map
upThree = list()
convLeftFlag = np.zeros((convLeft.shape[0], convLeft.shape[1]))
convRightFlag = np.zeros((convRight.shape[0], convRight.shape[1]))
for item in matchtPatchCoor:
if item[0] < 0 or item[1] < 0 or item[4] < 0 or item[5] < 0 or item[2] >= convLeft.shape[0] or item[3] >= \
convLeft.shape[1] or item[6] >= convRight.shape[0] or item[7] >= convRight.shape[1]:
continue
else:
currentPatchFilter, convLeftFlag, convRightFlag = pyramidRelu(item, convLeft, convRight, level, convLeftFlag, convRightFlag)
for ele in currentPatchFilter:
if ele not in upThree:
upThree.append(ele)
else:
continue
# upThree = pyramidRansac(upThree, convLeft, convRight, 14*(2**(5-level)), 4**(5-level))
# pyramidPlot(upThree, convLeft, convRight, 14 * (2 ** (5 - level)))
if level == 2 or level == 3:
upThree = pyramidRansac(upThree, convLeft, convRight, 14 * (2 ** (5 - level)), 200)
matchtUpPatchCoor = list()
# [LowApx, LowApy, highApx, highApy, LowBpx, LowBpy, highBpx, highBpy]
radius = 3 if level >= 4 else 2
if level == 1:
return upThree
for item in upThree:
# item: [Apx, Apy, Bpx, Bpy]
LowApx, HighApx = 2 * item[0] - radius, 2 * item[0] + radius
LowApy, HighApy = 2 * item[1] - radius, 2 * item[1] + radius
LowBpx, HighBpx = 2 * item[2] - radius, 2 * item[2] + radius
LowBpy, HighBpy = 2 * item[3] - radius, 2 * item[3] + radius
matchtUpPatchCoor.append([LowApx, LowApy, HighApx, HighApy, LowBpx, LowBpy, HighBpx, HighBpy])
print("pyramid " + str(level) + " end !\n")
print("The number of NBB matching points is : {}\n".format(len(matchtUpPatchCoor)))
print("The correspondence regions in Layer_" + str(level-1) + " (before filtering) is : {}\n".format(len(matchtUpPatchCoor)))
return upThree, matchtUpPatchCoor
def writeFilename(matchPoints, filename):
with open(filename, 'a') as f:
for item in matchPoints:
f.write(str(item[0]) + ',' + str(item[1]) + ',' + str(item[2]) + ',' + str(item[3]) + '\n')
def plotSameStyle(A, B):
assert A.shape[2] == B.shape[2]
meanA, meanB = np.zeros((1, A.shape[2])).squeeze(), np.zeros((1, B.shape[2])).squeeze()
stdA, stdB = np.zeros((1, A.shape[2])).squeeze(), np.zeros((1, B.shape[2])).squeeze()
for i in range(A.shape[2]):
meanA[i], stdA[i] = A[:, :, i].mean() + eps, A[:, :, i].std() + eps
meanB[i], stdB[i] = B[:, :, i].mean() + eps, B[:, :, i].std() + eps
meanPQ, stdPQ = (meanA + meanB) / 2, (stdA + stdB) / 2
# print(meanA, stdA)
CaPQ = ((A - meanA) / stdA) * stdPQ + meanPQ # size is row * col * Channel
CbQP = ((B - meanB) / stdB) * stdPQ + meanPQ # with size (row, col, Channel)
appendImage(CaPQ, CbQP) # plot the inter-same style image
def pyramidRansac(upFilter, reluFiveLeft, reluFiveRight, step, err):
imgTotalSource = appendImage(reluFiveLeft[:, :,0], reluFiveRight[:, :, 0])
# plt.imshow(imgTotalSource)
# flag = 0
# for item in upFilter:
# plt.plot(int(item[1]), int(item[0]), 'ro')
# plt.plot(int(item[3]) + step, int(item[2]), 'ro')
# plt.plot([int(item[1]), int(item[3]) + step], [int(item[0]), int(item[2])], color[flag % 8])
# flag += 1
#
# plt.colorbar()
# plt.axis('off')
# plt.show()
upFiveFilter, lenth, pointsNum, rightMatrix = plot_ransac_match(upFilter, step, imgTotalSource, err, [1, 1, 1, 1])
return upFiveFilter
def plot_Source(matchPoints, imgTotalSource, depth, imgASource, imgBSource):
# filename = 'NBB.txt'
# writeFilename(matchPoints, filename)
scale0A, scale1A = imgASource.shape[0] / depth, imgASource.shape[1] / depth # used for plot the source scale imageA
scale0B, scale1B = imgBSource.shape[0] / depth, imgBSource.shape[1] / depth # used for plot the source scale imageB
# plt.title("NBB with " + str(len(matchPoints)) + 'points')
plt.title(str(depth))
plt.imshow(imgTotalSource)
flag = 0
color = ['b', 'c', 'g', 'k', 'm', 'r', 'w', 'y']
for item in matchPoints:
plt.plot(int(item[1] * scale1A), int(item[0] * scale0A), 'r.')
plt.plot(int(item[3] * scale1B) + imgASource.shape[1], int(item[2] * scale0B), 'r.')
plt.plot([int(item[1] * scale1A), int(item[3] * scale1B) + imgASource.shape[1]], \
[int(item[0] * scale0A), int(item[2] * scale0B)], color[flag % 8])
item[1] = int(item[1] * scale1A)
item[0] = int(item[0] * scale0A)
item[3] = int(item[3] * scale1B)
item[2] = int(item[2] * scale0B)
flag += 1
plt.axis('off')
plt.savefig(str(depth) + 'real.png', dpi=300)
plt.show()