-
Notifications
You must be signed in to change notification settings - Fork 19
/
render.py
1935 lines (1646 loc) · 89.9 KB
/
render.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
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
import numpy as np
import pdb
#import cv2
import matplotlib.pyplot as plt
from rendering import filters
from skimage import io, color
import os
#from cv2.ximgproc import *
import scipy.interpolate as sinterp
def render_lens_imgs(lenses, lens_imgs, img_shape=None):
"""
Parameters
----------
lenses: dictionary, keys are integer pairs (axial hex coordinates)
The lens dictionary
lens_imgs: dictionary
Dictionary with the lens data, same size as lenses
img_shape: pair of integers
Shape of the target image
Returns
-------
img: array like
Two-dimensional array containing the microlens depth image
"""
assert len(lenses) == len(lens_imgs), "Number of lenses do not coincide"
assert len(lenses) > 0, "0 lenses supplied"
first_lens = lenses[0, 0]
# ensure that the center lens is at the image origin
if img_shape is None:
img_shape = ((first_lens.pcoord) * 2 + 1).astype(int)
# check if it is a colored image or a one-channel gray/disparity image
if len(lens_imgs[0,0].shape) == 3:
hl, wl, c = lens_imgs[0,0].shape
else:
hl, wl = lens_imgs[0,0].shape
c = 1
# here we create the structure for the image (circle images with the mask)
assert hl == wl
n = (hl - 1) / 2.0
x = np.linspace(-n, n, hl)
XX, YY = np.meshgrid(x, x)
ind = np.where(XX**2 + YY**2 < first_lens.inner_radius**2)
# micro image, so it takes the shape of first_lens.col_img_shape
if len(lens_imgs[0,0].shape) == 3:
img = np.zeros((img_shape[0], img_shape[1], c))
else:
img = np.zeros((img_shape))
#img = np.zeros((lens_imgs[0,0].shape))
for key in lenses:
data = np.asarray(lens_imgs[key])
lens = lenses[key]
ty = (YY + lens.pcoord[0] + 0.5).astype(int)
tx = (XX + lens.pcoord[1] + 0.5).astype(int)
# ensure that the subimg is located within the image bounds
if np.any(ty < 0) or np.any(tx < 0) or np.any(ty >= img_shape[0]) or np.any(tx >= img_shape[1]):
continue
if len(data.shape) > 0:
img[(ty[ind], tx[ind])] = data[ind]
else:
img[(ty[ind], tx[ind])] = data
return img
def render_cropped_img(lenses, lens_imgs, x1, y1, x2, y2):
"""
Parameters
----------
lenses: dictionary, keys are integer pairs (axial hex coordinates)
The lens dictionary
lens_imgs: dictionary
Dictionary with the lens data, same size as lenses
img_shape: pair of integers
Shape of the target image
Returns
-------
img: array like
Two-dimensional array containing the microlens depth image
"""
assert len(lenses) == len(lens_imgs), "Number of lenses do not coincide"
assert len(lenses) > 0, "0 lenses supplied"
first_lens = lenses[0, 0]
central_img = lens_imgs[0,0]
# ensure that the center lens is at the image origin
if img_shape is None:
img_shape = ((first_lens.pcoord) * 2 + 1).astype(int)
# check if it's gray image (disparity) or colored image
if len(central_img.shape) == 3:
hl, wl, c = central_img.shape
else:
hl, wl = central_img.shape
c = 1
# here we create the structure for the image (circle images with the mask)
assert hl == wl
n = (hl - 1) / 2.0
x = np.linspace(-n, n, hl)
XX, YY = np.meshgrid(x, x)
ind = np.where(XX**2 + YY**2 < first_lens.inner_radius**2)
# micro image, so it takes the shape of first_lens.col_img_shape
if len(central_img.shape) == 3:
img = np.zeros((img_shape[0], img_shape[1], c))
else:
img = np.zeros((img_shape))
for key in lenses:
#pdb.set_trace()
data = np.asarray(lens_imgs[key])
#l_type = ((-key[0] % 3) +key[1]) % 3
lens = lenses[key]
ty = (YY + lens.pcoord[0] + 0.5).astype(int)
tx = (XX + lens.pcoord[1] + 0.5).astype(int)
# ensure that the subimg is located within the image bounds
if np.any(ty < 0) or np.any(tx < 0) or np.any(ty >= img_shape[0]) or np.any(tx >= img_shape[1]):
continue
if len(data.shape) > 0:
img[(ty[ind], tx[ind])] = data[ind]
else:
img[(ty[ind], tx[ind])] = data
return img
def get_patch_size_fine(disp_img, min_d, max_d, max_ps, isReal=True, layers=3):
disparray = np.asarray(disp_img)
mean_d = np.mean(disparray)
std_d = np.std(disparray)
step = (max_d - min_d ) / layers
if isReal:
ps = max_ps - layers
for i in range(layers):
if mean_d > min_d + step * i:
ps += 1
else:
ps = max_ps
for i in range(layers):
if mean_d > min_d + step * i:
ps -= 1
return max(ps, 0)
"""
The idea is that if we can find the right parameters, the patch size should be
consistent across images.
We know that if we have the diameter of the lens, the disparity can reach
up to almost half of it, and minimum will be close to zero.
We also know that if for example disparity is close to zero, the patch size have to be really small
If disparity would be zero (focal plane case) then 1 pixel would be enough.
If disparity would be half of the lens diameter, the patch size should be close to half of the half of the diameters
so something like half o the disparity.
We just select some values in the middle also, dividing disparity in slices.
Also note that the patch size has to be odd (because of having one central pixel)
Later we can use a radius and select circular patches and then we have more levels
"""
def get_patch_size_absolute(disp_img, lens_diameter, isReal=True):
min_ps = 1
max_ps = np.floor(lens_diameter / 2)
if max_ps % 2 == 0:
max_ps += 1
number_of_different_sizes = (max_ps - min_ps) / 2 + 1
disparray = np.asarray(disp_img)
mean_d = np.mean(disparray) * max_ps
ps = np.ceil(mean_d * 0.5).astype(int)
if ps < 1:
ps = 1
#print("disp {0} and patch size {1}".format(mean_d, ps))
return ps
def get_patch_size_absolute_focused_lenses(disp_img, lens_diameter, isReal=True):
min_ps = 5
max_ps = np.floor(lens_diameter / 2)
if max_ps % 2 == 0:
max_ps += 1
number_of_different_sizes = (max_ps - min_ps) / 2 + 1
disparray = np.asarray(disp_img)
mean_d = np.mean(disparray) * max_ps
ps = np.round(max_ps * 0.4).astype(int)
#ps = np.round(max_ps - (mean_d)).astype(int)
if ps < 1:
ps = 1
#print("disp {0} and patch size {1}".format(mean_d, ps))
#ps = np.round(ps * 1.75).astype(int)
#if ps > max_ps:
# ps = max_ps
return ps
"""
REFOCUSING using patches of pixels from micro-images
or total focus also, depending on the use of the actual disparity
--------------
October 2018
"""
def refocused_using_patches(lenses, col_data, disp_data, min_disp, max_disp, max_ps=5, layers = 4, isReal=True, imgname=None):
if disp_data is None:
# refocusing!
# not ready yet
return None
# we set the patch image to be one fourth of the original, if not otherwise specified
factor = 4 # if changing this the final resolution will change
central_lens = lenses[0,0]
img_shape = ((central_lens.pcoord) * 2 + 1).astype(int)
cen = round(central_lens.img.shape[0]/2.0)
if len(col_data[0,0].shape) > 1:
hl, wl, c = col_data[0,0].shape
else:
hl, wl = central_lens.img.shape
c = 1
n = (hl - 1) / 2.0
x = np.linspace(-n, n, hl)
XX, YY = np.meshgrid(x, x)
ref_img = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor), c))
disp_ref_img = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
if c == 4:
ref_img[:,:,3] = 1 # alpha channel
count = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
psimg = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
actual_size = round(hl / factor)
if actual_size % 2 == 0:
actual_size += 1
dim = (actual_size, actual_size)
hw = int(np.floor(actual_size/2))
for key in lenses:
lens = lenses[key]
current_img = np.asarray(col_data[key])
current_disp = np.asarray(disp_data[key])
ps = get_patch_size_fine(current_disp, min_disp, max_disp, max_ps, isReal, layers)
cen_y, cen_x = int(round(lens.pcoord[0])), int(round(lens.pcoord[1]))
ptc_y, ptc_x = int(cen_y / factor), int(cen_x / factor)
if min(ptc_y, ptc_x) > max_ps and ptc_y < (ref_img.shape[0]-max_ps) and ptc_x < (ref_img.shape[1]-max_ps):
color_img = current_img[cen-ps:cen+ps+1, cen-ps:cen+ps+1] # patch size!
disp_simg = current_disp[cen-ps:cen+ps+1, cen-ps:cen+ps+1]
img_big = cv2.resize(color_img, dim, interpolation = cv2.INTER_LINEAR)
disp_big = cv2.resize(disp_simg, dim, interpolation = cv2.INTER_LINEAR)
count[ptc_y-hw:ptc_y+hw+1, ptc_x-hw:ptc_x+hw+1] += 1
psimg[ptc_y-hw:ptc_y+hw+1, ptc_x-hw:ptc_x+hw+1] = ps
ref_img[ptc_y-hw:ptc_y+hw+1, ptc_x-hw:ptc_x+hw+1, 0:3] += img_big[:,:,0:3]
disp_ref_img[ptc_y-hw:ptc_y+hw+1, ptc_x-hw:ptc_x+hw+1] += disp_big
ref_img_fnl = np.ones_like(ref_img)
disp_ref_img_fnl = np.ones_like(disp_ref_img)
for j in range(0,3):
ref_img_fnl[:,:,j] = ref_img[:,:,j] / count
disp_ref_img_fnl = disp_ref_img / count
return ref_img_fnl, disp_ref_img_fnl, psimg
def rgb2gray(rgb):
return np.dot(rgb[...,:3], [0.299, 0.587, 0.114])
"""
It creates a traditional image extracting patch from the lenslet image
Resolution is set to 1/4, still need to be updated to be changeable
Patch size is chosen automatically from disparity image
Using x_shift and y_shift is possible to obtain perspective shifts, i.e. different viewpoints
--------------
February 2019
"""
def generate_a_perspective_view(lenses, col_data, disp_data, min_disp, max_disp, x_shift=0, y_shift=0, cutBorders=True, isReal=True, imgname=None):
if disp_data is None:
# refocusing!
# not ready yet
return None
# we set the patch image to be one fourth of the original, if not otherwise specified
factor = 4 # if changing this the final resolution will change
central_lens = lenses[0,0]
img_shape = ((central_lens.pcoord) * 2 + 1).astype(int)
cen = round(central_lens.img.shape[0]/2.0)
if len(col_data[0,0].shape) > 1:
hl, wl, c = col_data[0,0].shape
else:
hl, wl = central_lens.img.shape
c = 1
max_ps = np.floor(central_lens.diameter / 2)
n = (hl - 1) / 2.0
x = np.linspace(-n, n, hl)
XX, YY = np.meshgrid(x, x)
ref_img = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor), c))
disp_ref_img = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
if c == 4:
ref_img[:,:,3] = 1 # alpha channel
count = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
psimg = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
actual_size = round(hl / factor)
if actual_size % 2 == 0:
actual_size += 1
dim = (actual_size, actual_size)
hw = int(np.floor(actual_size/2))
for key in lenses:
#pdb.set_trace()
lens = lenses[key]
current_img = np.asarray(col_data[key])
current_disp = np.asarray(disp_data[key])
ps = get_patch_size_absolute(current_disp, lens.diameter, isReal)
cen_y, cen_x = int(round(lens.pcoord[0])), int(round(lens.pcoord[1]))
ptc_y, ptc_x = int(cen_y / factor), int(cen_x / factor)
if min(ptc_y, ptc_x) > max_ps and ptc_y < (ref_img.shape[0]-max_ps) and ptc_x < (ref_img.shape[1]-max_ps):
color_img = current_img[cen-ps+y_shift:cen+ps+1+y_shift, cen-ps+x_shift:cen+ps+1+x_shift] # patch size!
disp_simg = current_disp[cen-ps+y_shift:cen+ps+1+y_shift, cen-ps+x_shift:cen+ps+1+x_shift]
#pdb.set_trace()
#print("size of color_img {0}".format(color_img.shape))
#test_img = current_img[cen-ps:cen+ps+1, cen-ps:cen+ps+1]
#print("size without shift {0}".format(test_img.shape))
img_big = cv2.resize(color_img, dim, interpolation = cv2.INTER_LINEAR)
disp_big = cv2.resize(disp_simg, dim, interpolation = cv2.INTER_LINEAR)
count[ptc_y-hw:ptc_y+hw+1, ptc_x-hw:ptc_x+hw+1] += 1
psimg[ptc_y-hw:ptc_y+hw+1, ptc_x-hw:ptc_x+hw+1] = ps #color_img.shape[0] * color_img.shape[1]
ref_img[ptc_y-hw:ptc_y+hw+1, ptc_x-hw:ptc_x+hw+1, 0:3] += img_big[:,:,0:3]
disp_ref_img[ptc_y-hw:ptc_y+hw+1, ptc_x-hw:ptc_x+hw+1] += disp_big
ref_img_fnl = np.ones_like(ref_img)
disp_ref_img_fnl = np.ones_like(disp_ref_img)
count[(count == 0)] = 1
for j in range(0,3):
ref_img_fnl[:,:,j] = ref_img[:,:,j] / count
disp_ref_img_fnl = disp_ref_img / count
ref_img_fnl[np.isnan(ref_img_fnl)] = 0
disp_ref_img_fnl[np.isnan(disp_ref_img_fnl)] = 0
if cutBorders is True:
paddingToAvoidBorders = int(max_ps + 1)
ref_img_fnl = ref_img_fnl[paddingToAvoidBorders:ref_img_fnl.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:ref_img_fnl.shape[1]-paddingToAvoidBorders, :]
disp_ref_img_fnl = disp_ref_img_fnl[paddingToAvoidBorders:disp_ref_img_fnl.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:disp_ref_img_fnl.shape[1]-paddingToAvoidBorders]
psimg = psimg[paddingToAvoidBorders:psimg.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:psimg.shape[1]-paddingToAvoidBorders]
return ref_img_fnl, disp_ref_img_fnl, psimg
### It just generates three images (usually colored image, disparity and confidence
### but can be used with whatever is loaded with load_triplet)
def generate_a_perspective_view_triplet(lenses, x_shift=0, y_shift=0, cutBorders=True, isReal=True, imgname=None):
# we set the patch image to be one fourth of the original, if not otherwise specified
factor = 4 # if changing this the final resolution will change
central_lens = lenses[0,0]
img_shape = ((central_lens.pcoord) * 2 + 1).astype(int)
cen = round(central_lens.img.shape[0]/2.0)
if len(lenses[0,0].col_img.shape) > 1:
hl, wl, c = lenses[0,0].col_img.shape
else:
hl, wl = central_lens.img.shape
c = 1
max_ps = np.floor(central_lens.diameter / 2)
n = (hl - 1) / 2.0
x = np.linspace(-n, n, hl)
XX, YY = np.meshgrid(x, x)
ref_img = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor), c))
# we assume they don't have colors! (usually disp and confidence do not have channels..)
ref_disp = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
ref_conf = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
if c == 4:
ref_img[:,:,3] = 1 # alpha channel
count = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
actual_size = round(hl / factor)
if actual_size % 2 == 0:
actual_size += 1
dim = (actual_size, actual_size)
hw = int(np.floor(actual_size/2))
for key in lenses:
#pdb.set_trace()
lens = lenses[key]
current_img = np.asarray(lens.col_img)
current_disp = np.asarray(lens.disp_img)
current_conf = np.asarray(lens.conf_img)
ps = get_patch_size_absolute(current_disp, lens.diameter, isReal)
cen_y, cen_x = int(round(lens.pcoord[0])), int(round(lens.pcoord[1]))
ptc_y, ptc_x = int(cen_y / factor), int(cen_x / factor)
if min(ptc_y, ptc_x) > max_ps and ptc_y < (ref_img.shape[0]-max_ps) and ptc_x < (ref_img.shape[1]-max_ps):
color_img = current_img[cen-ps+y_shift:cen+ps+1+y_shift, cen-ps+x_shift:cen+ps+1+x_shift] # patch size!
disp_simg = current_disp[cen-ps+y_shift:cen+ps+1+y_shift, cen-ps+x_shift:cen+ps+1+x_shift]
conf_simg = current_conf[cen-ps+y_shift:cen+ps+1+y_shift, cen-ps+x_shift:cen+ps+1+x_shift]#pdb.set_trace()
#print("size of color_img {0}".format(color_img.shape))
#test_img = current_img[cen-ps:cen+ps+1, cen-ps:cen+ps+1]
#print("size without shift {0}".format(test_img.shape))
img_big = cv2.resize(color_img, dim, interpolation = cv2.INTER_LINEAR)
disp_big = cv2.resize(disp_simg, dim, interpolation = cv2.INTER_LINEAR)
conf_big = cv2.resize(conf_simg, dim, interpolation = cv2.INTER_LINEAR)
count[ptc_y-hw:ptc_y+hw+1, ptc_x-hw:ptc_x+hw+1] += 1
ref_img[ptc_y-hw:ptc_y+hw+1, ptc_x-hw:ptc_x+hw+1, 0:3] += img_big[:,:,0:3]
ref_disp[ptc_y-hw:ptc_y+hw+1, ptc_x-hw:ptc_x+hw+1] += disp_big
ref_conf[ptc_y-hw:ptc_y+hw+1, ptc_x-hw:ptc_x+hw+1] += conf_big
ref_img_fnl = np.ones_like(ref_img)
#disp_ref_img_fnl = np.ones_like(disp_ref_img)
count[(count == 0)] = 1
for j in range(0,3):
ref_img_fnl[:,:,j] = ref_img[:,:,j] / count
disp_ref_img_fnl = ref_disp / count
conf_ref_img_fnl = ref_conf / count
ref_img_fnl[np.isnan(ref_img_fnl)] = 0
disp_ref_img_fnl[np.isnan(disp_ref_img_fnl)] = 0
conf_ref_img_fnl[np.isnan(conf_ref_img_fnl)] = 0
if cutBorders is True:
paddingToAvoidBorders = int(max_ps + 1)
ref_img_fnl = ref_img_fnl[paddingToAvoidBorders:ref_img_fnl.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:ref_img_fnl.shape[1]-paddingToAvoidBorders, :]
disp_ref_img_fnl = disp_ref_img_fnl[paddingToAvoidBorders:disp_ref_img_fnl.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:disp_ref_img_fnl.shape[1]-paddingToAvoidBorders]
conf_ref_img_fnl = conf_ref_img_fnl[paddingToAvoidBorders:conf_ref_img_fnl.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:conf_ref_img_fnl.shape[1]-paddingToAvoidBorders]
#psimg = psimg[paddingToAvoidBorders:psimg.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:psimg.shape[1]-paddingToAvoidBorders]
return ref_img_fnl, disp_ref_img_fnl, conf_ref_img_fnl
"""
Createas a view using only micro-lenses that are on focus
doing so, spatial resolution is reduced but also blur and artifacts.
It first creates three images using only one lens type, then pick the part of
those images that are in focus and merge them together using a weighted average
the idea is that by averaging them together you reduce artifacts (in shiny parts and edges)
but by using weights (so weight more the ones that are in focus) you keep the sharpness
--------------
February 2019
"""
def generate_view_focused_micro_lenses(lenses, min_disp=0, max_disp=0, no_conf=False, x_shift=0, y_shift=0, patch_shape=0, cutBorders=True, isReal=True, imgname=None):
# bilateral filter
triplet = [[12, 5, 7], [10, 7, 9], [8, 11, 13], [6, 13, 15], [4, 15, 17]]
chosen = 3
# we set the patch image to be one/eigth of the original, if not otherwise specified
factor = triplet[chosen][0] # if changing this the final resolution will change
central_lens = lenses[0,0]
if max_disp == 0:
max_disp = central_lens.diameter
img_shape = ((central_lens.pcoord) * 2 + 1).astype(int)
cen = round(central_lens.img.shape[0]/2.0)
if len(central_lens.col_img.shape) > 1:
hl, wl, c = central_lens.col_img.shape
else:
hl, wl = central_lens.img.shape
c = 1
max_ps = np.floor(central_lens.diameter / 2)
img_lens_type0 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor), c))
img_lens_type1 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor), c))
img_lens_type2 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor), c))
if c == 4:
img_lens_type0[:,:,3] = 1 # alpha channel
img_lens_type1[:,:,3] = 1 # alpha channel
img_lens_type2[:,:,3] = 1 # alpha channel
disp_lens_type0 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
disp_lens_type1 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
disp_lens_type2 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
if no_conf == False:
conf_lens_type0 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
conf_lens_type1 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
conf_lens_type2 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
count0 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
count1 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
count2 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
psimg0 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
psimg1 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
psimg2 = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor)))
actual_size_x = triplet[chosen][1] #15
actual_size_y = triplet[chosen][2] #round(hl / factor) + 4
if actual_size_x % 2 == 0:
actual_size_x += 1
dim = (actual_size_x, actual_size_y)
hw_x = int(np.floor(actual_size_x/2))
hw_y = int(np.floor(actual_size_y/2))
# create a mask to actual extract eclipses patches
radius = np.floor(actual_size_y/2)
x = np.linspace(-1, 1, actual_size_y) * radius
xx, yy = np.meshgrid(x, x)
if patch_shape == 0:
rect_mask = np.ones_like(xx)
mask = rect_mask[:,1:rect_mask.shape[1]-1]
elif patch_shape == 1:
circle_mask = np.zeros_like(xx)
circle_mask[xx**2 + yy**2 < (radius+1)**2] = 1
mask = circle_mask[:,1:circle_mask.shape[1]-1]
mask4c = np.dstack((mask, mask, mask, mask))
# loop and create three images!
for key in lenses:
#pdb.set_trace()
lens = lenses[key]
current_img = np.asarray(col_data[key])
current_disp = np.asarray(disp_data[key])
if no_conf == False:
current_conf = np.asarray(conf_data[key])
ps = get_patch_size_absolute_focused_lenses(current_disp, lens.diameter, isReal)
cen_y, cen_x = int(np.round(lens.pcoord[0])), int(np.floor(lens.pcoord[1]))
ptc_y, ptc_x = int(cen_y / factor), int(cen_x / factor)
if min(ptc_y, ptc_x) > max_ps and ptc_y < (img_lens_type0.shape[0]-max_ps) and ptc_x < (img_lens_type0.shape[1]-max_ps):
color_img = current_img[cen-ps+y_shift:cen+ps+1+y_shift, cen-ps+x_shift:cen+ps+1+x_shift] # patch size!
disp_simg = current_disp[cen-ps+y_shift:cen+ps+1+y_shift, cen-ps+x_shift:cen+ps+1+x_shift]
if no_conf == False:
conf_img = current_conf[cen-ps+y_shift:cen+ps+1+y_shift, cen-ps+x_shift:cen+ps+1+x_shift]
img_big = cv2.resize(color_img, dim, interpolation = cv2.INTER_LINEAR) * mask4c
disp_big = cv2.resize(disp_simg, dim, interpolation = cv2.INTER_LINEAR) * mask
if no_conf == False:
conf_big = cv2.resize(conf_img, dim, interpolation = cv2.INTER_LINEAR) * mask
if lens.focal_type == 0:
count0[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1] += mask
psimg0[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1] = mask * ps#color_img.shape[0] * color_img.shape[1]
img_lens_type0[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1, 0:3] += img_big[:,:,0:3]
disp_lens_type0[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1] += disp_big
if no_conf == False:
conf_lens_type0[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1] += conf_big
elif lens.focal_type == 1:
count1[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1] += mask
psimg1[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1] = mask * ps#color_img.shape[0] * color_img.shape[1]
img_lens_type1[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1, 0:3] += img_big[:,:,0:3]
disp_lens_type1[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1] += disp_big
if no_conf == False:
conf_lens_type1[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1] += conf_big
elif lens.focal_type == 2:
count2[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1] += mask
psimg2[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1] = mask * ps#color_img.shape[0] * color_img.shape[1]
img_lens_type2[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1, 0:3] += img_big[:,:,0:3]
disp_lens_type2[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1] += disp_big
if no_conf == False:
conf_lens_type2[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1] += conf_big
# Here I should average the three images, but first get them right
# yes, terribly written, but is temporary I hope
img_lens_type0_fnl = np.ones_like(img_lens_type0)
img_lens_type1_fnl = np.ones_like(img_lens_type1)
img_lens_type2_fnl = np.ones_like(img_lens_type2)
disp_lens_type0_fnl = np.ones_like(disp_lens_type0)
disp_lens_type1_fnl = np.ones_like(disp_lens_type1)
disp_lens_type2_fnl = np.ones_like(disp_lens_type2)
if no_conf == False:
conf_lens_type0_fnl = np.ones_like(conf_lens_type0)
conf_lens_type1_fnl = np.ones_like(conf_lens_type1)
conf_lens_type2_fnl = np.ones_like(conf_lens_type2)
count0[(count0 == 0)] = 1
count1[(count1 == 0)] = 1
count2[(count2 == 0)] = 1
for j in range(0,3):
img_lens_type0_fnl[:,:,j] = img_lens_type0[:,:,j] / count0
img_lens_type1_fnl[:,:,j] = img_lens_type1[:,:,j] / count1
img_lens_type2_fnl[:,:,j] = img_lens_type2[:,:,j] / count2
disp_lens_type0_fnl = disp_lens_type0 / count0
disp_lens_type1_fnl = disp_lens_type1 / count1
disp_lens_type2_fnl = disp_lens_type2 / count2
if no_conf == False:
conf_lens_type0_fnl = conf_lens_type0 / count0
conf_lens_type1_fnl = conf_lens_type1 / count1
conf_lens_type2_fnl = conf_lens_type2 / count2
img_lens_type0_fnl[np.isnan(img_lens_type0_fnl)] = 0
img_lens_type1_fnl[np.isnan(img_lens_type1_fnl)] = 0
img_lens_type2_fnl[np.isnan(img_lens_type2_fnl)] = 0
disp_lens_type0_fnl[np.isnan(disp_lens_type0_fnl)] = 0
disp_lens_type1_fnl[np.isnan(disp_lens_type1_fnl)] = 0
disp_lens_type2_fnl[np.isnan(disp_lens_type2_fnl)] = 0
if no_conf == False:
conf_lens_type0_fnl[np.isnan(disp_lens_type0_fnl)] = 0
conf_lens_type1_fnl[np.isnan(disp_lens_type1_fnl)] = 0
conf_lens_type2_fnl[np.isnan(disp_lens_type2_fnl)] = 0
# select disparity
avg_disp = (disp_lens_type0_fnl + disp_lens_type1_fnl + disp_lens_type2_fnl) / 3
# divide areas
# lens type 0 --> 1 to 3 virtual depth --> disparity > 0.6
# lens type 1 --> 3 to 4 virtual depth --> 0.6 > disparity > 0.3
# lens type 2 --> 4 to 100 virtual depth --> disparity < 0.3
weights = np.zeros((img_lens_type0_fnl.shape[0], img_lens_type0_fnl.shape[1], 4))
lens_type0_focus_area = avg_disp > 0.6
lens_type1_focus_area = (avg_disp > 0.3) * (avg_disp < 0.6)
lens_type2_focus_area = avg_disp < 0.3
weights[:,:,0] = 0.6 * lens_type0_focus_area + 0.2 * lens_type1_focus_area + 0.1 * lens_type2_focus_area
weights[:,:,1] = 0.3 * lens_type0_focus_area + 0.6 * lens_type1_focus_area + 0.3 * lens_type2_focus_area
weights[:,:,2] = 0.1 * lens_type0_focus_area + 0.2 * lens_type1_focus_area + 0.6 * lens_type2_focus_area
weights[:,:,3] = np.ones_like(weights[:,:,3])
all_in_focus_image = (img_lens_type0_fnl * np.dstack((weights[:,:,0], weights[:,:,0], weights[:,:,0], weights[:,:,3])) + \
img_lens_type1_fnl * np.dstack((weights[:,:,1], weights[:,:,1], weights[:,:,1], weights[:,:,3])) + \
img_lens_type2_fnl * np.dstack((weights[:,:,2], weights[:,:,2], weights[:,:,2], weights[:,:,3])) )
all_in_focus_image[:,:,3] = 1
final_disp_img = (disp_lens_type0_fnl * weights[:,:,0] + disp_lens_type1_fnl * weights[:,:,1] + disp_lens_type2_fnl * weights[:,:,2] )
if no_conf == False:
final_conf_img = (conf_lens_type0_fnl * weights[:,:,0] + conf_lens_type1_fnl * weights[:,:,1] + conf_lens_type2_fnl * weights[:,:,2] )
else:
final_conf_img = np.zeros_like(final_disp_img)
avg_ps = (psimg0 + psimg1 + psimg2 ) / 3
# cutting out the sides where there is no information!
if cutBorders is True:
paddingToAvoidBorders = int(max_ps + 1)
all_in_focus_image = all_in_focus_image[paddingToAvoidBorders:all_in_focus_image.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:all_in_focus_image.shape[1]-paddingToAvoidBorders, :]
final_disp_img = final_disp_img[paddingToAvoidBorders:final_disp_img.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:final_disp_img.shape[1]-paddingToAvoidBorders]
avg_disp = avg_disp[paddingToAvoidBorders:avg_disp.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:avg_disp.shape[1]-paddingToAvoidBorders]
avg_ps = avg_ps[paddingToAvoidBorders:avg_ps.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:avg_ps.shape[1]-paddingToAvoidBorders]
final_conf_img = final_conf_img[paddingToAvoidBorders:final_conf_img.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:final_conf_img.shape[1]-paddingToAvoidBorders]
return all_in_focus_image, avg_disp, final_disp_img, avg_ps, final_conf_img
"""
GENERATE FOCUSED VIEW
Rewritten for better implemetnation
adding filters in the end
April 2019
"""
def generate_view_focused_micro_lenses_v2(lenses, no_conf=False, x_shift=0, y_shift=0, patch_shape=0, cutBorders=True, isReal=True, imgname=None, chosen=3):
# patch shapes
# they should be related to the microlens image size (these numbers were good for R29)
triplet = [[12, 5, 7], [10, 7, 9], [8, 11, 13], [6, 13, 15], [4, 15, 17]]
lens_size = lenses[0,0].diameter
#triplet = [[12, 5, 7], [10, 7, 9], [8, 11, 13], [6, 23, 25], [4, 51, 55]]
chosen = 1
# for the images captured with 25mm objective
# triplet25mm = [6, 23, 25] # lens diameter 70 pixels
# for the old images
# triplet25mm = [6, 13, 13] # lens diameter 40 pixels
lens_types = 3
# we set the patch image to be one/sixth of the original, if not otherwise specified
factor = triplet[chosen][0] # if changing this the final resolution will change
central_lens = lenses[0,0]
img_shape = ((central_lens.pcoord) * 2 + 1).astype(int)
cen = round(central_lens.img.shape[0]/2.0)
if len(central_lens.col_img.shape) > 1:
hl, wl, c = central_lens.col_img.shape
else:
hl, wl = central_lens.img.shape
c = 1
max_ps = np.floor(central_lens.diameter / 2)
# WE USE 3 DIMENSIONAL STRUCTURES
# They basically are layer of images (with color channels, or with one value for disparity and confidence).
# Each layer has information from one lens type
rendered_colors = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor), c, lens_types))
if c == 4:
rendered_colors[:,:,3,:] = 1 # alpha channel
rendered_disps = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor), lens_types))
if no_conf == False:
rendered_confidences = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor), lens_types))
counters = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor), lens_types))
patch_sizes = np.zeros((int(img_shape[0]/factor), int(img_shape[1]/factor), lens_types))
# actual size of the patches in the rendered image
actual_size_x = triplet[chosen][1] #15
actual_size_y = triplet[chosen][2] #round(hl / factor) + 4
if actual_size_x % 2 == 0:
actual_size_x += 1
dim = (actual_size_x, actual_size_y)
hw_x = int(np.floor(actual_size_x/2))
hw_y = int(np.floor(actual_size_y/2))
# create a mask to actual extract eclipses patches
radius = np.floor(actual_size_y/2)
x = np.linspace(-1, 1, actual_size_y) * radius
xx, yy = np.meshgrid(x, x)
if patch_shape == 0:
rect_mask = np.ones_like(xx)
mask = rect_mask[:,2:rect_mask.shape[1]]
# for the big microlenses
#mask = rect_mask[:,2:rect_mask.shape[1]-2]
elif patch_shape == 1:
circle_mask = np.zeros_like(xx)
circle_mask[xx**2 + yy**2 < (radius+1)**2] = 1
mask = rect_mask[:,2:rect_mask.shape[1]]
# for the big microlenses
#mask = rect_mask[:,2:rect_mask.shape[1]-2]
mask4c = np.dstack((mask, mask, mask, mask))
# loop and create three images!
for key in lenses:
#pdb.set_trace()
lens = lenses[key]
current_img = np.asarray(lenses[key].col_img)
current_disp = np.asarray(lenses[key].disp_img)
if no_conf == False:
current_conf = np.asarray(lenses[key].conf_img)
ps = get_patch_size_absolute_focused_lenses(current_disp, lens.diameter, isReal)
#print(ps)
cen_y, cen_x = int(np.round(lens.pcoord[0])), int(np.floor(lens.pcoord[1]))
ptc_y, ptc_x = int(cen_y / factor), int(cen_x / factor)
if min(ptc_y, ptc_x) > max_ps and ptc_y < (rendered_colors.shape[0]-max_ps) and ptc_x < (rendered_colors.shape[1]-max_ps):
#pdb.set_trace()
color_img = current_img[cen-ps+y_shift:cen+ps+1+y_shift, cen-ps+x_shift:cen+ps+1+x_shift] # patch size!
disp_simg = current_disp[cen-ps+y_shift:cen+ps+1+y_shift, cen-ps+x_shift:cen+ps+1+x_shift]
if no_conf == False:
conf_img = current_conf[cen-ps+y_shift:cen+ps+1+y_shift, cen-ps+x_shift:cen+ps+1+x_shift]
img_big = cv2.resize(color_img, dim, interpolation = cv2.INTER_LINEAR) * mask4c
disp_big = cv2.resize(disp_simg, dim, interpolation = cv2.INTER_LINEAR) * mask
if no_conf == False:
conf_big = cv2.resize(conf_img, dim, interpolation = cv2.INTER_LINEAR) * mask
# using lens.focal_type we fill only one of the three layers each time
rendered_colors[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1,0:3, lens.focal_type] += img_big[:,:,0:3]
rendered_disps[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1, lens.focal_type] += disp_big
counters[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1, lens.focal_type] += mask
patch_sizes[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1, lens.focal_type] = mask * ps
if no_conf == False:
rendered_confidences[ptc_y-hw_y:ptc_y+hw_y+1, ptc_x-hw_x:ptc_x+hw_x+1, lens.focal_type] += conf_big
counters[(counters == 0)] = 1
#pdb.set_trace()
# each lens type has to be divided
for k in range(0,3):
# each color channel
for j in range(0,3):
rendered_colors[:,:,j,k] /= counters[:,:,k]
rendered_disps[:,:,k] /= counters[:,:,k]
#rendered_disps[np.isnan(rendered_disps[:,:,k]), k] = 0
if no_conf == False:
rendered_confidences[:,:,k] /= counters[:,:,k]
#rendered_confidences[np.isnan(rendered_confidences[:,:,k]), k] = 0
initial_disp = np.mean(rendered_disps, axis=2)
average_patch_sizes = np.mean(patch_sizes, axis=2)
# divide areas
# lens type 0 --> 1 to 3 virtual depth --> disparity > 0.6
# lens type 1 --> 3 to 4 virtual depth --> 0.6 > disparity > 0.3
# lens type 2 --> 4 to 100 virtual depth --> disparity < 0.3
weights = np.zeros((initial_disp.shape[0], initial_disp.shape[1], 4))
lens_type0_focus_area = initial_disp > 0.6
lens_type1_focus_area = (initial_disp > 0.3) * (initial_disp < 0.6)
lens_type2_focus_area = initial_disp < 0.3
#pdb.set_trace()
# here we need a bette rway
weights[:,:,0] = 0.625#0.6 * lens_type0_focus_area + 0.2 * lens_type1_focus_area + 0.1 * lens_type2_focus_area
weights[:,:,1] = 0.05# * lens_type0_focus_area + 0.6 * lens_type1_focus_area + 0.3 * lens_type2_focus_area
weights[:,:,2] = 0.325# * lens_type0_focus_area + 0.2 * lens_type1_focus_area + 0.6 * lens_type2_focus_area
weights[:,:,3] = np.ones_like(weights[:,:,3])
all_in_focus_image = np.zeros_like(rendered_colors[:,:,:,0])
final_disp_img = np.zeros_like(rendered_disps[:,:,0])
final_conf_img = np.zeros_like(final_disp_img)
#pdb.set_trace()
for s in range(0,3):
all_in_focus_image += (rendered_colors[:,:,:,s] * np.dstack((weights[:,:,s], weights[:,:,s], weights[:,:,s], weights[:,:,3])))
final_disp_img += (rendered_disps[:,:,s] * weights[:,:,s])
if no_conf == False:
final_conf_img += (rendered_confidences[:,:,s] * weights[:,:,s])
all_in_focus_image[:,:,3] = 1
### FILTERING
# the color image is filtered with a soft bilateral filter (standard settings)
window_size = 13
sigma_distance = 0.75
sigma_color = 0.5
print("Processing colored image..")
all_in_focus_image = filters.bilateral_filter(all_in_focus_image, window_size, sigma_distance, sigma_color)
print("Processing disparity map..")
# sigmaSpatial = 3
# sigmaColor = 3
#pdb.set_trace()
processed_disp = filters.median_filter(final_disp_img, window_size)
#bf = filters.bilateral_filter(final_disp_img, 13, 1)
#gf = guidedFilter(all_in_focus_image,final_disp_img.astype(np.float32), 13, 0.5)
#jbf = jointBilateralFilter(all_in_focus_image, final_disp_img.astype(np.float32), 9, sigmaColor, sigmaSpatial)
#dtf = dtFilter(all_in_focus_image, final_disp_img.astype(np.float32), sigmaSpatial, sigmaColor)
#processed_disp = filters.replace_wrong_values(final_disp_img, all_in_focus_image, final_conf_img, minDensity = 0.5)
#processed_disp2 = filters.replace_wrong_values(final_disp_img, all_in_focus_image, final_conf_img, minDensity = 0.75)
#processed_disp3 = filters.replace_wrong_values(final_disp_img, all_in_focus_image, final_conf_img, minDensity = 0.25)
# plt.ion()
# #plt.imshow(final_disp_img)
# plt.figure(1); plt.imshow(final_disp_img, cmap='jet', vmin=0.25, vmax= 0.75)
# plt.figure(2); plt.imshow(mf, cmap='jet', vmin=0.25, vmax= 0.75)
# #plt.figure(3); plt.imshow(bf, cmap='jet', vmin=0.25, vmax= 0.75)
# #plt.figure(4); plt.imshow(gf, cmap='jet', vmin=0.25, vmax= 0.75)
# #plt.figure(4); plt.imshow(jbf, cmap='jet', vmin=0.25, vmax= 0.75)
# plt.figure(5); plt.imshow(dtf, cmap='jet', vmin=0.25, vmax= 0.75)
# #plt.show()
# pdb.set_trace()
# cutting out the sides where there is no information!
if cutBorders is True:
paddingToAvoidBorders = int(max_ps + 1)
all_in_focus_image = all_in_focus_image[paddingToAvoidBorders:all_in_focus_image.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:all_in_focus_image.shape[1]-paddingToAvoidBorders, :]
final_disp_img = final_disp_img[paddingToAvoidBorders:final_disp_img.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:final_disp_img.shape[1]-paddingToAvoidBorders]
initial_disp = initial_disp[paddingToAvoidBorders:initial_disp.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:initial_disp.shape[1]-paddingToAvoidBorders]
average_patch_sizes = average_patch_sizes[paddingToAvoidBorders:average_patch_sizes.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:average_patch_sizes.shape[1]-paddingToAvoidBorders]
final_conf_img = final_conf_img[paddingToAvoidBorders:final_conf_img.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:final_conf_img.shape[1]-paddingToAvoidBorders]
processed_disp = processed_disp[paddingToAvoidBorders:processed_disp.shape[0]-paddingToAvoidBorders, paddingToAvoidBorders:processed_disp.shape[1]-paddingToAvoidBorders]
return all_in_focus_image, initial_disp, final_disp_img, average_patch_sizes, final_conf_img, processed_disp
def get_sampling_distance(disp, calib, sam_per_lens):
disp_in_pixel = disp * calib.lens_diameter
if disp_in_pixel < 0.001:
disp_in_pixel = 0.5
sam_dist = disp_in_pixel / (2 * sam_per_lens)
return sam_dist
def _hex_focal_type(c):
"""
Calculates the focal type for the three lens hexagonal grid
"""
focal_type = ((-c[0] % 3) + c[1]) % 3
return focal_type
def render_interp_img(imgs, interps, calibs, shiftx, shifty, cut_borders):
img = imgs[0]
disp = imgs[1]
data_interp_r = interps[0]
data_interp_g = interps[1]
data_interp_b = interps[2]
disp_interp = interps[3]
# view
img_shape = np.asarray(img.shape[0:2])
calib = calibs[0]
coords = calibs[1]
local_grid = calibs[2]
# resolution should be correlated with number of lenses more than
# number of pixels
#pdb.set_trace()
# sample per lens
sam_per_lens = 11
hs = np.floor(sam_per_lens/2).astype(int)
#[ny * sam_per_lens, nx * sam_per_lens]
#pdb.set_trace()
reducing_factor = (calib.lens_diameter / sam_per_lens)
resolution = np.round(img_shape / reducing_factor).astype(int)
print("raw image is {}x{}, rendered image will be {}x{}".format(img_shape[0], img_shape[1], resolution[0], resolution[1]))
rnd_img = np.zeros((resolution[0], resolution[1], 3))
x, y = local_grid.x, local_grid.y
# if needed for masking
# xx, yy = local_grid.xx, local_grid.yy
# mask = np.zeros_like(local_grid.xx)
# mask[xx**2 + yy**2 < calib.inner_lens_radius**2] = 1
#plt.ion()
for lc in coords:
# pixel coordinates
pc = coords[lc]
#pdb.set_trace()
#print("pc[0] = {}, pc[1] = {}".format(pc[0], pc[1]))
# first we need disparity
disp_at_pc = disp_interp(y+pc[0], x+pc[1])
# we need a single value for the disparity
single_val_disp = np.mean(disp_at_pc)
# disparity controls distance between pixels
sampling_distance = get_sampling_distance(single_val_disp, calib, sam_per_lens)
#print("disp is {}, sampling distance is {}".format(single_val_disp * calib.lens_diameter, sampling_distance))
# sample the image at the correct position
#pdb.set_trace()
coords_resized = pc / reducing_factor
intPCx = np.ceil(coords_resized[1]).astype(int)
intPCy = np.ceil(coords_resized[0]).astype(int)
if intPCx > hs and resolution[1] - intPCx > hs and intPCy > hs and resolution[0] - intPCy > hs:
sampling_pattern = np.arange(-sampling_distance*sam_per_lens/2, sampling_distance*sam_per_lens/2 + sampling_distance, sampling_distance)
sampling_pattern_x = sampling_pattern + shiftx
sampling_pattern_y = sampling_pattern + shifty
patch_values = np.dstack((data_interp_r(sampling_pattern_y+pc[0], sampling_pattern_x+pc[1]),
data_interp_g(sampling_pattern_y+pc[0], sampling_pattern_x+pc[1]),
data_interp_b(sampling_pattern_y+pc[0], sampling_pattern_x+pc[1])))
patch_values = np.clip(patch_values, 0, np.max(patch_values))
#print("patch_values size {}".format(patch_values.shape))
interp_patch_r = sinterp.RectBivariateSpline(range(patch_values.shape[0]), range(patch_values.shape[1]), patch_values[:,:,0])
interp_patch_g = sinterp.RectBivariateSpline(range(patch_values.shape[0]), range(patch_values.shape[1]), patch_values[:,:,1])
interp_patch_b = sinterp.RectBivariateSpline(range(patch_values.shape[0]), range(patch_values.shape[1]), patch_values[:,:,2])
sampling_pattern_for_patch_y = np.arange((intPCy-coords_resized[0]), (intPCy-coords_resized[0]+sam_per_lens), 1)
sampling_pattern_for_patch_x = np.arange((intPCx-coords_resized[1]), (intPCx-coords_resized[1]+sam_per_lens), 1)
r_channel = interp_patch_r(sampling_pattern_for_patch_y, sampling_pattern_for_patch_x)
g_channel = interp_patch_g(sampling_pattern_for_patch_y, sampling_pattern_for_patch_x)
b_channel = interp_patch_b(sampling_pattern_for_patch_y, sampling_pattern_for_patch_x)
rgb_interp_patch_img = np.dstack((r_channel, g_channel, b_channel))
rgb_interp_patch_img = np.clip(rgb_interp_patch_img, 0, np.max(rgb_interp_patch_img))
#print("sam patt y {}, x {}".format(sampling_pattern_for_patch_y.shape, sampling_pattern_for_patch_x.shape))
#print("size {}:{}, {}:{}".format(intPCy-hs,intPCy+hs+1, intPCx-hs,intPCx+hs+1))
# pdb.set_trace()
# plt.figure(1)
# plt.subplot(131)
# plt.imshow(patch_values)
# plt.subplot(132)
# cX = int(round(pc[1]))
# cY = int(round(pc[0]))
# plt.imshow(img[cY-15:cY+16, cX-15:cX+16])
# plt.subplot(133)
# plt.imshow((rgb_interp_patch_img))
#pdb.set_trace()
rnd_img[intPCy-hs:intPCy+hs+1, intPCx-hs:intPCx+hs+1,:] = rgb_interp_patch_img
# rnd_img[intPCy-hs:intPCy+hs+1, intPCx-hs:intPCx+hs+1,1] = interp_patch_g(sampling_pattern_for_patch_y, sampling_pattern_for_patch_x)
# rnd_img[intPCy-hs:intPCy+hs+1, intPCx-hs:intPCx+hs+1,2] = interp_patch_b(sampling_pattern_for_patch_y, sampling_pattern_for_patch_x)
rnd_img = np.clip(rnd_img, 0, 1)
if cut_borders:
rnd_img = rnd_img[hs:rnd_img.shape[0]-hs, hs:rnd_img.shape[1]-hs,:]
#plt.ion()
#plt.imshow((rnd_img))
#pdb.set_trace()
#plt.imsave('/data1/palmieri/COLLABORATIONS/Waqas/IMAGES/RAYTRIX/OUTPUT/RTX008/interp2.png', np.clip(rnd_img, 0, 1))
return rnd_img
def render_interp_img_focused(imgs, interps, calibs, shiftx, shifty, sam_per_lens, cut_borders):
img = imgs[0]
disp = imgs[1]
data_interp_r = interps[0]
data_interp_g = interps[1]
data_interp_b = interps[2]
disp_interp = interps[3]
# view
img_shape = np.asarray(img.shape[0:2])