/
dic.cu
893 lines (778 loc) · 71.6 KB
/
dic.cu
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
#include "dic.h"
#include "dicimage.h"
#include "cuda.h"
#include "cuda_runtime.h"
#include "cuda_runtime_api.h"
#include "device_launch_parameters.h"
// Define this to turn on error checking
#define CUDA_ERROR_CHECK
#define cudaCheckError() __cudaCheckError(__FILE__, __LINE__)
inline void __cudaCheckError(const char *file, const int line) {
#ifdef CUDA_ERROR_CHECK
cudaError err = cudaGetLastError();
if (cudaSuccess != err) {
fprintf(stderr, "cudaCheckError() failed at %s:%i : %s\n",
file, line, cudaGetErrorString(err));
exit(-1);
}
// More careful checking. However, this will affect performance.
// Comment away if needed.
// err = cudaDeviceSynchronize();
// if(cudaSuccess != err) {
// fprintf(stderr, "cudaCheckError() with sync failed at %s:%i : %s\n",
// file, line, cudaGetErrorString( err ));
// exit(-1);
// }
#endif
return;
}
Dic::Dic() {}
Dic::~Dic() {
cimgs.clear();
free(df_dx_buffer);
free(df_dy_buffer);
df_dp_buffer.clear();
g_buffer.clear();
x_vec_buffer.clear();
y_vec_buffer.clear();
gradient_buffer.clear();
hessian_gn_buffer.clear();
QK_B_QKT_buffer.clear();
for (std::size_t i = 0; i < cimgs.size(); i++) {
plot_u[i].free();
plot_v[i].free();
plot_corrcoef[i].free();
}
plot_calcpoints.reset();
}
void Dic::performDicAnalysis() {
std::ofstream of("./dic_analysis.txt");
double start = clock();
int height = rimg.gs.height;
int width = rimg.gs.width;
oHeight = static_cast<int>(ceil(static_cast<double>(rimg.gs.height) / static_cast<double>(params.subsetSpacing + 1)));
oWidth = static_cast<int>(ceil(static_cast<double>(rimg.gs.width) / static_cast<double>(params.subsetSpacing + 1)));
df_dx_buffer = (double*) malloc(height * width * sizeof(double));
df_dx_buffer = (double*) malloc(height * width * sizeof(double));
int radius = params.subsetSize;
plot_calcpoints.alloc(oHeight, oWidth);
g_buffer.resize((radius * 2 + 1) * (radius * 2 + 1), 0.0);
df_dp_buffer.resize((radius * 2 + 1) * (radius * 2 + 1) * 6, 0.0);
x_vec_buffer.resize(6, 0.0);
y_vec_buffer.resize(6, 0.0);
gradient_buffer.resize(6, 0.0);
hessian_gn_buffer.resize(36, 0.0);
QK_B_QKT_buffer.resize(36 * (rimg.bcoef.height - 5) * (rimg.bcoef.width - 5), 0.0);
for (std::size_t i = 0; i < cimgs.size(); i++) {
class_double_array plotu;
class_double_array plotv;
class_double_array plotcc;
class_logical_array plotvp;
plotu.alloc(oHeight, oWidth);
plotv.alloc(oHeight, oWidth);
plotcc.alloc(oHeight, oWidth);
plotvp.alloc(oHeight, oWidth);
plot_u.push_back(plotu);
plot_v.push_back(plotv);
plot_corrcoef.push_back(plotcc);
plot_validpoints.push_back(plotvp);
}
bool result = preComputeRef();
if(!result) {
printf("Error in precompute");
return;
}
/*// Dic for each current image
for (std::size_t i = 0; i < cimgs.size(); i++) {
preCompute(of, i);
roi.set_cirroi(radius);
analysis(of, i);
plot_calcpoints.reset();
}
of.close();*/
}
__global__
void pcrKernel(double* df_dx_buffer, double* df_dy_buffer,
int f_height, int f_width,
double* bcoef, int b_height, int border_bcoef,
int offset) {
//bounds
int li = border_bcoef - 2;
int ri = border_bcoef + f_width - 1;
int lj = border_bcoef - 2;
int rj = border_bcoef + f_height - 1;
int rows = (border_bcoef + f_width - 1) - (border_bcoef - 2) + 1;
int cols = (border_bcoef + f_height - 1) - (border_bcoef - 2) + 1;
//init with 0 based indexing, then add border_bcoef info
int i = (offset + threadIdx.x) / cols;
int j = (offset + threadIdx.x) % cols;
//give proper final value
i += li;
j += lj;
if(i > ri || j > rj) return;
double b0 = bcoef[(j) + (i) * b_height];
double b1 = bcoef[(j + 1) + (i) * b_height];
double b2 = bcoef[(j + 2) + (i) * b_height];
double b3 = bcoef[(j + 3) + (i) * b_height];
double b4 = bcoef[(j + 4) + (i) * b_height];
double b6 = bcoef[(j) + (i + 1) * b_height];
double b7 = bcoef[(j + 1) + (i + 1) * b_height];
double b8 = bcoef[(j + 2) + (i + 1) * b_height];
double b9 = bcoef[(j + 3) + (i + 1) * b_height];
double b10 = bcoef[(j + 5) + (i + 1) * b_height];
double b12 = bcoef[(j) + (i + 2) * b_height];
double b13 = bcoef[(j + 1) + (i + 2) * b_height];
double b15 = bcoef[(j + 3) + (i + 2) * b_height];
double b16 = bcoef[(j + 4) + (i + 2) * b_height];
double b18 = bcoef[(j) + (i + 3) * b_height];
double b19 = bcoef[(j + 1) + (i + 3) * b_height];
double b20 = bcoef[(j + 2) + (i + 3) * b_height];
double b21 = bcoef[(j + 3) + (i + 3) * b_height];
double b22 = bcoef[(j + 4) + (i + 3) * b_height];
double b24 = bcoef[(j) + (i + 4) * b_height];
double b25 = bcoef[(j + 1) + (i + 4) * b_height];
double b26 = bcoef[(j + 2) + (i + 4) * b_height];
double b27 = bcoef[(j + 3) + (i + 4) * b_height];
double b28 = bcoef[(j + 4) + (i + 4) * b_height];
// Compute base index
std::size_t lind_f = (j - (border_bcoef - 2)) + (i - (border_bcoef - 2)) * f_height;
// Compute Gradients using b-spline coefficients
// First order
df_dx_buffer[lind_f] = 0.003472222222222222 * b18 - 0.009027777777777778 * b1 - 0.003472222222222222 * b10 - 0.0003472222222222222 * b0 + 0.09027777777777778 * b19 - 0.02291666666666667 * b2 + 0.2291666666666667 * b20 + 0.09027777777777778 * b21 + 0.003472222222222222 * b22 + 0.0003472222222222222 * b24 + 0.009027777777777778 * b25 + 0.02291666666666667 * b26 + 0.009027777777777778 * b27 + 0.0003472222222222222 * b28 - 0.009027777777777778 * b3 - 0.0003472222222222222 * b4 - 0.003472222222222222 * b6 - 0.09027777777777778 * b7 - 0.2291666666666667 * b8 - 0.09027777777777778 * b9;
df_dy_buffer[lind_f] = 0.009027777777777778 * b10 - 0.003472222222222222 * b1 - 0.0003472222222222222 * b0 - 0.02291666666666667 * b12 - 0.2291666666666667 * b13 + 0.2291666666666667 * b15 + 0.02291666666666667 * b16 - 0.009027777777777778 * b18 - 0.09027777777777778 * b19 + 0.09027777777777778 * b21 + 0.009027777777777778 * b22 - 0.0003472222222222222 * b24 - 0.003472222222222222 * b25 + 0.003472222222222222 * b27 + 0.0003472222222222222 * b28 + 0.003472222222222222 * b3 + 0.0003472222222222222 * b4 - 0.009027777777777778 * b6 - 0.09027777777777778 * b7 + 0.09027777777777778 * b9;
}
bool Dic::preComputeRef() {
//copy reference image to device
//stl is not directly supported in cuda
printf("Precomputing reference image\n");
double* device_df_dx_buffer;
double* device_df_dy_buffer;
int size = rimg.gs.height * rimg.gs.width;
//allocate memory in device
if(cudaSuccess != cudaMalloc(&device_df_dx_buffer, size * sizeof(double))) return false;
if(cudaSuccess != cudaMalloc(&device_df_dy_buffer, size * sizeof(double))) return false;
//see https://devtalk.nvidia.com/default/topic/465306/using-std-vector-in-cuda-kernel-its-posible-to-use-a-std-vector-inside-cuda-kernel-/
//if(cudaSuccess != cudaMemcpy(device_df_dx_buffer, &df_dx_buffer[0], size * sizeof(double), cudaMemcpyHostToDevice)) return false;
//if(cudaSuccess != cudaMemcpy(device_df_dy_buffer, &df_dy_buffer[0], size * sizeof(double), cudaMemcpyHostToDevice)) return false;
double* device_rimg_bcoef_value;
//top level class malloc
int bcoefSize = (rimg.gs.height + 2 * rimg.border_bcoef) * (rimg.gs.width + 2 * rimg.border_bcoef);
if(cudaSuccess != cudaMalloc(&device_rimg_bcoef_value, bcoefSize * sizeof(double))) return false;
//memcpy class_img
if(cudaSuccess != cudaMemcpy(device_rimg_bcoef_value, rimg.bcoef.value,
bcoefSize * sizeof(double), cudaMemcpyHostToDevice)) return false;
int threads = 512;
int rows = (rimg.border_bcoef + rimg.gs.width - 1) - (rimg.border_bcoef - 2) + 1;
int cols = (rimg.border_bcoef + rimg.gs.height - 1) - (rimg.border_bcoef - 2) + 1;
int pSize = rows * cols;
int iterations = pSize / threads + (pSize % threads != 0);
for(int i = 0, offset = 0; i < iterations; i++, offset += threads) {
pcrKernel<<<1, threads>>>(device_df_dx_buffer, device_df_dy_buffer,
rimg.gs.height, rimg.gs.width,
device_rimg_bcoef_value, rimg.bcoef.height, rimg.border_bcoef,
offset);
}
cudaDeviceSynchronize();
cudaCheckError();
//get back results
if(cudaSuccess != cudaMemcpy(df_dx_buffer, device_df_dx_buffer, size * sizeof(double), cudaMemcpyDeviceToHost)) return false;
if(cudaSuccess != cudaMemcpy(df_dy_buffer, device_df_dy_buffer, size * sizeof(double), cudaMemcpyDeviceToHost)) return false;
//free resources
cudaFree(device_df_dx_buffer);
cudaFree(device_df_dy_buffer);
cudaFree(device_rimg_bcoef_value);
return true;
}
void Dic::preCompute(std::ofstream &of, std::size_t currImg) {
// Pre compute interpolation coefficients ---------------------------//
int height = cimgs[currImg].bcoef.height;
for (int i = 0; i < cimgs[currImg].bcoef.width - 5; i++) {
for (int j = 0; j < height - 5; j++) {
// Get bspline coefficients
double b0 = cimgs[currImg].bcoef.value[(j) + (i) * height];
double b1 = cimgs[currImg].bcoef.value[(j + 1) + (i) * height];
double b2 = cimgs[currImg].bcoef.value[(j + 2) + (i) * height];
double b3 = cimgs[currImg].bcoef.value[(j + 3) + (i) * height];
double b4 = cimgs[currImg].bcoef.value[(j + 4) + (i) * height];
double b5 = cimgs[currImg].bcoef.value[(j + 5) + (i) * height];
double b6 = cimgs[currImg].bcoef.value[(j) + (i + 1) * height];
double b7 = cimgs[currImg].bcoef.value[(j + 1) + (i + 1) * height];
double b8 = cimgs[currImg].bcoef.value[(j + 2) + (i + 1) * height];
double b9 = cimgs[currImg].bcoef.value[(j + 3) + (i + 1) * height];
double b10 = cimgs[currImg].bcoef.value[(j + 4) + (i + 1) * height];
double b11 = cimgs[currImg].bcoef.value[(j + 5) + (i + 1) * height];
double b12 = cimgs[currImg].bcoef.value[(j) + (i + 2) * height];
double b13 = cimgs[currImg].bcoef.value[(j + 1) + (i + 2) * height];
double b14 = cimgs[currImg].bcoef.value[(j + 2) + (i + 2) * height];
double b15 = cimgs[currImg].bcoef.value[(j + 3) + (i + 2) * height];
double b16 = cimgs[currImg].bcoef.value[(j + 4) + (i + 2) * height];
double b17 = cimgs[currImg].bcoef.value[(j + 5) + (i + 2) * height];
double b18 = cimgs[currImg].bcoef.value[(j) + (i + 3) * height];
double b19 = cimgs[currImg].bcoef.value[(j + 1) + (i + 3) * height];
double b20 = cimgs[currImg].bcoef.value[(j + 2) + (i + 3) * height];
double b21 = cimgs[currImg].bcoef.value[(j + 3) + (i + 3) * height];
double b22 = cimgs[currImg].bcoef.value[(j + 4) + (i + 3) * height];
double b23 = cimgs[currImg].bcoef.value[(j + 5) + (i + 3) * height];
double b24 = cimgs[currImg].bcoef.value[(j) + (i + 4) * height];
double b25 = cimgs[currImg].bcoef.value[(j + 1) + (i + 4) * height];
double b26 = cimgs[currImg].bcoef.value[(j + 2) + (i + 4) * height];
double b27 = cimgs[currImg].bcoef.value[(j + 3) + (i + 4) * height];
double b28 = cimgs[currImg].bcoef.value[(j + 4) + (i + 4) * height];
double b29 = cimgs[currImg].bcoef.value[(j + 5) + (i + 4) * height];
double b30 = cimgs[currImg].bcoef.value[(j) + (i + 5) * height];
double b31 = cimgs[currImg].bcoef.value[(j + 1) + (i + 5) * height];
double b32 = cimgs[currImg].bcoef.value[(j + 2) + (i + 5) * height];
double b33 = cimgs[currImg].bcoef.value[(j + 3) + (i + 5) * height];
double b34 = cimgs[currImg].bcoef.value[(j + 4) + (i + 5) * height];
double b35 = cimgs[currImg].bcoef.value[(j + 5) + (i + 5) * height];
//Compute base index
int lind_qkbqkt = j * (36) + i * (36 * (height - 5));
//Compute QK_B_QKT vector
QK_B_QKT_buffer[lind_qkbqkt] = 0.00006944444444444444 * b0 + 0.001805555555555556 * b1 + 0.001805555555555556 * b10 + 0.004583333333333333 * b12 + 0.1191666666666667 * b13 + 0.3025 * b14 + 0.1191666666666667 * b15 + 0.004583333333333333 * b16 + 0.001805555555555556 * b18 + 0.04694444444444444 * b19 + 0.004583333333333333 * b2 + 0.1191666666666667 * b20 + 0.04694444444444444 * b21 + 0.001805555555555556 * b22 + 0.00006944444444444444 * b24 + 0.001805555555555556 * b25 + 0.004583333333333333 * b26 + 0.001805555555555556 * b27 + 0.00006944444444444444 * b28 + 0.001805555555555556 * b3 + 0.00006944444444444444 * b4 + 0.001805555555555556 * b6 + 0.04694444444444444 * b7 + 0.1191666666666667 * b8 + 0.04694444444444444 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 1] = 0.009027777777777778 * b10 - 0.003472222222222222 * b1 - 0.0003472222222222222 * b0 - 0.02291666666666667 * b12 - 0.2291666666666667 * b13 + 0.2291666666666667 * b15 + 0.02291666666666667 * b16 - 0.009027777777777778 * b18 - 0.09027777777777778 * b19 + 0.09027777777777778 * b21 + 0.009027777777777778 * b22 - 0.0003472222222222222 * b24 - 0.003472222222222222 * b25 + 0.003472222222222222 * b27 + 0.0003472222222222222 * b28 + 0.003472222222222222 * b3 + 0.0003472222222222222 * b4 - 0.009027777777777778 * b6 - 0.09027777777777778 * b7 + 0.09027777777777778 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 2] = 0.0006944444444444444 * b0 + 0.001388888888888889 * b1 + 0.01805555555555556 * b10 + 0.04583333333333333 * b12 + 0.09166666666666667 * b13 - 0.275 * b14 + 0.09166666666666667 * b15 + 0.04583333333333333 * b16 + 0.01805555555555556 * b18 + 0.03611111111111111 * b19 - 0.004166666666666667 * b2 - 0.1083333333333333 * b20 + 0.03611111111111111 * b21 + 0.01805555555555556 * b22 + 0.0006944444444444444 * b24 + 0.001388888888888889 * b25 - 0.004166666666666667 * b26 + 0.001388888888888889 * b27 + 0.0006944444444444444 * b28 + 0.001388888888888889 * b3 + 0.0006944444444444444 * b4 + 0.01805555555555556 * b6 + 0.03611111111111111 * b7 - 0.1083333333333333 * b8 + 0.03611111111111111 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 3] = 0.001388888888888889 * b1 - 0.0006944444444444444 * b0 + 0.01805555555555556 * b10 - 0.04583333333333333 * b12 + 0.09166666666666667 * b13 - 0.09166666666666667 * b15 + 0.04583333333333333 * b16 - 0.01805555555555556 * b18 + 0.03611111111111111 * b19 - 0.03611111111111111 * b21 + 0.01805555555555556 * b22 - 0.0006944444444444444 * b24 + 0.001388888888888889 * b25 - 0.001388888888888889 * b27 + 0.0006944444444444444 * b28 - 0.001388888888888889 * b3 + 0.0006944444444444444 * b4 - 0.01805555555555556 * b6 + 0.03611111111111111 * b7 - 0.03611111111111111 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 4] = 0.0003472222222222222 * b0 - 0.001388888888888889 * b1 + 0.009027777777777778 * b10 + 0.02291666666666667 * b12 - 0.09166666666666667 * b13 + 0.1375 * b14 - 0.09166666666666667 * b15 + 0.02291666666666667 * b16 + 0.009027777777777778 * b18 - 0.03611111111111111 * b19 + 0.002083333333333333 * b2 + 0.05416666666666667 * b20 - 0.03611111111111111 * b21 + 0.009027777777777778 * b22 + 0.0003472222222222222 * b24 - 0.001388888888888889 * b25 + 0.002083333333333333 * b26 - 0.001388888888888889 * b27 + 0.0003472222222222222 * b28 - 0.001388888888888889 * b3 + 0.0003472222222222222 * b4 + 0.009027777777777778 * b6 - 0.03611111111111111 * b7 + 0.05416666666666667 * b8 - 0.03611111111111111 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 5] = 0.0003472222222222222 * b1 - 0.00006944444444444444 * b0 - 0.009027777777777778 * b10 + 0.001805555555555556 * b11 - 0.004583333333333333 * b12 + 0.02291666666666667 * b13 - 0.04583333333333333 * b14 + 0.04583333333333333 * b15 - 0.02291666666666667 * b16 + 0.004583333333333333 * b17 - 0.001805555555555556 * b18 + 0.009027777777777778 * b19 - 0.0006944444444444444 * b2 - 0.01805555555555556 * b20 + 0.01805555555555556 * b21 - 0.009027777777777778 * b22 + 0.001805555555555556 * b23 - 0.00006944444444444444 * b24 + 0.0003472222222222222 * b25 - 0.0006944444444444444 * b26 + 0.0006944444444444444 * b27 - 0.0003472222222222222 * b28 + 0.00006944444444444444 * b29 + 0.0006944444444444444 * b3 - 0.0003472222222222222 * b4 + 0.00006944444444444444 * b5 - 0.001805555555555556 * b6 + 0.009027777777777778 * b7 - 0.01805555555555556 * b8 + 0.01805555555555556 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 6] = 0.003472222222222222 * b18 - 0.009027777777777778 * b1 - 0.003472222222222222 * b10 - 0.0003472222222222222 * b0 + 0.09027777777777778 * b19 - 0.02291666666666667 * b2 + 0.2291666666666667 * b20 + 0.09027777777777778 * b21 + 0.003472222222222222 * b22 + 0.0003472222222222222 * b24 + 0.009027777777777778 * b25 + 0.02291666666666667 * b26 + 0.009027777777777778 * b27 + 0.0003472222222222222 * b28 - 0.009027777777777778 * b3 - 0.0003472222222222222 * b4 - 0.003472222222222222 * b6 - 0.09027777777777778 * b7 - 0.2291666666666667 * b8 - 0.09027777777777778 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 7] = 0.001736111111111111 * b0 + 0.01736111111111111 * b1 - 0.01736111111111111 * b10 - 0.01736111111111111 * b18 - 0.1736111111111111 * b19 + 0.1736111111111111 * b21 + 0.01736111111111111 * b22 - 0.001736111111111111 * b24 - 0.01736111111111111 * b25 + 0.01736111111111111 * b27 + 0.001736111111111111 * b28 - 0.01736111111111111 * b3 - 0.001736111111111111 * b4 + 0.01736111111111111 * b6 + 0.1736111111111111 * b7 - 0.1736111111111111 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 8] = 0.03472222222222222 * b18 - 0.006944444444444444 * b1 - 0.03472222222222222 * b10 - 0.003472222222222222 * b0 + 0.06944444444444444 * b19 + 0.02083333333333333 * b2 - 0.2083333333333333 * b20 + 0.06944444444444444 * b21 + 0.03472222222222222 * b22 + 0.003472222222222222 * b24 + 0.006944444444444444 * b25 - 0.02083333333333333 * b26 + 0.006944444444444444 * b27 + 0.003472222222222222 * b28 - 0.006944444444444444 * b3 - 0.003472222222222222 * b4 - 0.03472222222222222 * b6 - 0.06944444444444444 * b7 + 0.2083333333333333 * b8 - 0.06944444444444444 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 9] = 0.003472222222222222 * b0 - 0.006944444444444444 * b1 - 0.03472222222222222 * b10 - 0.03472222222222222 * b18 + 0.06944444444444444 * b19 - 0.06944444444444444 * b21 + 0.03472222222222222 * b22 - 0.003472222222222222 * b24 + 0.006944444444444444 * b25 - 0.006944444444444444 * b27 + 0.003472222222222222 * b28 + 0.006944444444444444 * b3 - 0.003472222222222222 * b4 + 0.03472222222222222 * b6 - 0.06944444444444444 * b7 + 0.06944444444444444 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 10] = 0.006944444444444444 * b1 - 0.001736111111111111 * b0 - 0.01736111111111111 * b10 + 0.01736111111111111 * b18 - 0.06944444444444444 * b19 - 0.01041666666666667 * b2 + 0.1041666666666667 * b20 - 0.06944444444444444 * b21 + 0.01736111111111111 * b22 + 0.001736111111111111 * b24 - 0.006944444444444444 * b25 + 0.01041666666666667 * b26 - 0.006944444444444444 * b27 + 0.001736111111111111 * b28 + 0.006944444444444444 * b3 - 0.001736111111111111 * b4 - 0.01736111111111111 * b6 + 0.06944444444444444 * b7 - 0.1041666666666667 * b8 + 0.06944444444444444 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 11] = 0.0003472222222222222 * b0 - 0.001736111111111111 * b1 + 0.01736111111111111 * b10 - 0.003472222222222222 * b11 - 0.003472222222222222 * b18 + 0.01736111111111111 * b19 + 0.003472222222222222 * b2 - 0.03472222222222222 * b20 + 0.03472222222222222 * b21 - 0.01736111111111111 * b22 + 0.003472222222222222 * b23 - 0.0003472222222222222 * b24 + 0.001736111111111111 * b25 - 0.003472222222222222 * b26 + 0.003472222222222222 * b27 - 0.001736111111111111 * b28 + 0.0003472222222222222 * b29 - 0.003472222222222222 * b3 + 0.001736111111111111 * b4 - 0.0003472222222222222 * b5 + 0.003472222222222222 * b6 - 0.01736111111111111 * b7 + 0.03472222222222222 * b8 - 0.03472222222222222 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 12] = 0.0006944444444444444 * b0 + 0.01805555555555556 * b1 + 0.001388888888888889 * b10 - 0.004166666666666667 * b12 - 0.1083333333333333 * b13 - 0.275 * b14 - 0.1083333333333333 * b15 - 0.004166666666666667 * b16 + 0.001388888888888889 * b18 + 0.03611111111111111 * b19 + 0.04583333333333333 * b2 + 0.09166666666666667 * b20 + 0.03611111111111111 * b21 + 0.001388888888888889 * b22 + 0.0006944444444444444 * b24 + 0.01805555555555556 * b25 + 0.04583333333333333 * b26 + 0.01805555555555556 * b27 + 0.0006944444444444444 * b28 + 0.01805555555555556 * b3 + 0.0006944444444444444 * b4 + 0.001388888888888889 * b6 + 0.03611111111111111 * b7 + 0.09166666666666667 * b8 + 0.03611111111111111 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 13] = 0.006944444444444444 * b10 - 0.03472222222222222 * b1 - 0.003472222222222222 * b0 + 0.02083333333333333 * b12 + 0.2083333333333333 * b13 - 0.2083333333333333 * b15 - 0.02083333333333333 * b16 - 0.006944444444444444 * b18 - 0.06944444444444444 * b19 + 0.06944444444444444 * b21 + 0.006944444444444444 * b22 - 0.003472222222222222 * b24 - 0.03472222222222222 * b25 + 0.03472222222222222 * b27 + 0.003472222222222222 * b28 + 0.03472222222222222 * b3 + 0.003472222222222222 * b4 - 0.006944444444444444 * b6 - 0.06944444444444444 * b7 + 0.06944444444444444 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 14] = 0.006944444444444444 * b0 + 0.01388888888888889 * b1 + 0.01388888888888889 * b10 - 0.04166666666666667 * b12 - 0.08333333333333333 * b13 + 0.25 * b14 - 0.08333333333333333 * b15 - 0.04166666666666667 * b16 + 0.01388888888888889 * b18 + 0.02777777777777778 * b19 - 0.04166666666666667 * b2 - 0.08333333333333333 * b20 + 0.02777777777777778 * b21 + 0.01388888888888889 * b22 + 0.006944444444444444 * b24 + 0.01388888888888889 * b25 - 0.04166666666666667 * b26 + 0.01388888888888889 * b27 + 0.006944444444444444 * b28 + 0.01388888888888889 * b3 + 0.006944444444444444 * b4 + 0.01388888888888889 * b6 + 0.02777777777777778 * b7 - 0.08333333333333333 * b8 + 0.02777777777777778 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 15] = 0.01388888888888889 * b1 - 0.006944444444444444 * b0 + 0.01388888888888889 * b10 + 0.04166666666666667 * b12 - 0.08333333333333333 * b13 + 0.08333333333333333 * b15 - 0.04166666666666667 * b16 - 0.01388888888888889 * b18 + 0.02777777777777778 * b19 - 0.02777777777777778 * b21 + 0.01388888888888889 * b22 - 0.006944444444444444 * b24 + 0.01388888888888889 * b25 - 0.01388888888888889 * b27 + 0.006944444444444444 * b28 - 0.01388888888888889 * b3 + 0.006944444444444444 * b4 - 0.01388888888888889 * b6 + 0.02777777777777778 * b7 - 0.02777777777777778 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 16] = 0.003472222222222222 * b0 - 0.01388888888888889 * b1 + 0.006944444444444444 * b10 - 0.02083333333333333 * b12 + 0.08333333333333333 * b13 - 0.125 * b14 + 0.08333333333333333 * b15 - 0.02083333333333333 * b16 + 0.006944444444444444 * b18 - 0.02777777777777778 * b19 + 0.02083333333333333 * b2 + 0.04166666666666667 * b20 - 0.02777777777777778 * b21 + 0.006944444444444444 * b22 + 0.003472222222222222 * b24 - 0.01388888888888889 * b25 + 0.02083333333333333 * b26 - 0.01388888888888889 * b27 + 0.003472222222222222 * b28 - 0.01388888888888889 * b3 + 0.003472222222222222 * b4 + 0.006944444444444444 * b6 - 0.02777777777777778 * b7 + 0.04166666666666667 * b8 - 0.02777777777777778 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 17] = 0.003472222222222222 * b1 - 0.0006944444444444444 * b0 - 0.006944444444444444 * b10 + 0.001388888888888889 * b11 + 0.004166666666666667 * b12 - 0.02083333333333333 * b13 + 0.04166666666666667 * b14 - 0.04166666666666667 * b15 + 0.02083333333333333 * b16 - 0.004166666666666667 * b17 - 0.001388888888888889 * b18 + 0.006944444444444444 * b19 - 0.006944444444444444 * b2 - 0.01388888888888889 * b20 + 0.01388888888888889 * b21 - 0.006944444444444444 * b22 + 0.001388888888888889 * b23 - 0.0006944444444444444 * b24 + 0.003472222222222222 * b25 - 0.006944444444444444 * b26 + 0.006944444444444444 * b27 - 0.003472222222222222 * b28 + 0.0006944444444444444 * b29 + 0.006944444444444444 * b3 - 0.003472222222222222 * b4 + 0.0006944444444444444 * b5 - 0.001388888888888889 * b6 + 0.006944444444444444 * b7 - 0.01388888888888889 * b8 + 0.01388888888888889 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 18] = 0.001388888888888889 * b10 - 0.01805555555555556 * b1 - 0.0006944444444444444 * b0 - 0.001388888888888889 * b18 - 0.03611111111111111 * b19 - 0.04583333333333333 * b2 - 0.09166666666666667 * b20 - 0.03611111111111111 * b21 - 0.001388888888888889 * b22 + 0.0006944444444444444 * b24 + 0.01805555555555556 * b25 + 0.04583333333333333 * b26 + 0.01805555555555556 * b27 + 0.0006944444444444444 * b28 - 0.01805555555555556 * b3 - 0.0006944444444444444 * b4 + 0.001388888888888889 * b6 + 0.03611111111111111 * b7 + 0.09166666666666667 * b8 + 0.03611111111111111 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 19] = 0.003472222222222222 * b0 + 0.03472222222222222 * b1 + 0.006944444444444444 * b10 + 0.006944444444444444 * b18 + 0.06944444444444444 * b19 - 0.06944444444444444 * b21 - 0.006944444444444444 * b22 - 0.003472222222222222 * b24 - 0.03472222222222222 * b25 + 0.03472222222222222 * b27 + 0.003472222222222222 * b28 - 0.03472222222222222 * b3 - 0.003472222222222222 * b4 - 0.006944444444444444 * b6 - 0.06944444444444444 * b7 + 0.06944444444444444 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 20] = 0.01388888888888889 * b10 - 0.01388888888888889 * b1 - 0.006944444444444444 * b0 - 0.01388888888888889 * b18 - 0.02777777777777778 * b19 + 0.04166666666666667 * b2 + 0.08333333333333333 * b20 - 0.02777777777777778 * b21 - 0.01388888888888889 * b22 + 0.006944444444444444 * b24 + 0.01388888888888889 * b25 - 0.04166666666666667 * b26 + 0.01388888888888889 * b27 + 0.006944444444444444 * b28 - 0.01388888888888889 * b3 - 0.006944444444444444 * b4 + 0.01388888888888889 * b6 + 0.02777777777777778 * b7 - 0.08333333333333333 * b8 + 0.02777777777777778 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 21] = 0.006944444444444444 * b0 - 0.01388888888888889 * b1 + 0.01388888888888889 * b10 + 0.01388888888888889 * b18 - 0.02777777777777778 * b19 + 0.02777777777777778 * b21 - 0.01388888888888889 * b22 - 0.006944444444444444 * b24 + 0.01388888888888889 * b25 - 0.01388888888888889 * b27 + 0.006944444444444444 * b28 + 0.01388888888888889 * b3 - 0.006944444444444444 * b4 - 0.01388888888888889 * b6 + 0.02777777777777778 * b7 - 0.02777777777777778 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 22] = 0.01388888888888889 * b1 - 0.003472222222222222 * b0 + 0.006944444444444444 * b10 - 0.006944444444444444 * b18 + 0.02777777777777778 * b19 - 0.02083333333333333 * b2 - 0.04166666666666667 * b20 + 0.02777777777777778 * b21 - 0.006944444444444444 * b22 + 0.003472222222222222 * b24 - 0.01388888888888889 * b25 + 0.02083333333333333 * b26 - 0.01388888888888889 * b27 + 0.003472222222222222 * b28 + 0.01388888888888889 * b3 - 0.003472222222222222 * b4 + 0.006944444444444444 * b6 - 0.02777777777777778 * b7 + 0.04166666666666667 * b8 - 0.02777777777777778 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 23] = 0.0006944444444444444 * b0 - 0.003472222222222222 * b1 - 0.006944444444444444 * b10 + 0.001388888888888889 * b11 + 0.001388888888888889 * b18 - 0.006944444444444444 * b19 + 0.006944444444444444 * b2 + 0.01388888888888889 * b20 - 0.01388888888888889 * b21 + 0.006944444444444444 * b22 - 0.001388888888888889 * b23 - 0.0006944444444444444 * b24 + 0.003472222222222222 * b25 - 0.006944444444444444 * b26 + 0.006944444444444444 * b27 - 0.003472222222222222 * b28 + 0.0006944444444444444 * b29 - 0.006944444444444444 * b3 + 0.003472222222222222 * b4 - 0.0006944444444444444 * b5 - 0.001388888888888889 * b6 + 0.006944444444444444 * b7 - 0.01388888888888889 * b8 + 0.01388888888888889 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 24] = 0.0003472222222222222 * b0 + 0.009027777777777778 * b1 - 0.001388888888888889 * b10 + 0.002083333333333333 * b12 + 0.05416666666666667 * b13 + 0.1375 * b14 + 0.05416666666666667 * b15 + 0.002083333333333333 * b16 - 0.001388888888888889 * b18 - 0.03611111111111111 * b19 + 0.02291666666666667 * b2 - 0.09166666666666667 * b20 - 0.03611111111111111 * b21 - 0.001388888888888889 * b22 + 0.0003472222222222222 * b24 + 0.009027777777777778 * b25 + 0.02291666666666667 * b26 + 0.009027777777777778 * b27 + 0.0003472222222222222 * b28 + 0.009027777777777778 * b3 + 0.0003472222222222222 * b4 - 0.001388888888888889 * b6 - 0.03611111111111111 * b7 - 0.09166666666666667 * b8 - 0.03611111111111111 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 25] = 0.1041666666666667 * b15 - 0.01736111111111111 * b1 - 0.006944444444444444 * b10 - 0.01041666666666667 * b12 - 0.1041666666666667 * b13 - 0.001736111111111111 * b0 + 0.01041666666666667 * b16 + 0.006944444444444444 * b18 + 0.06944444444444444 * b19 - 0.06944444444444444 * b21 - 0.006944444444444444 * b22 - 0.001736111111111111 * b24 - 0.01736111111111111 * b25 + 0.01736111111111111 * b27 + 0.001736111111111111 * b28 + 0.01736111111111111 * b3 + 0.001736111111111111 * b4 + 0.006944444444444444 * b6 + 0.06944444444444444 * b7 - 0.06944444444444444 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 26] = 0.003472222222222222 * b0 + 0.006944444444444444 * b1 - 0.01388888888888889 * b10 + 0.02083333333333333 * b12 + 0.04166666666666667 * b13 - 0.125 * b14 + 0.04166666666666667 * b15 + 0.02083333333333333 * b16 - 0.01388888888888889 * b18 - 0.02777777777777778 * b19 - 0.02083333333333333 * b2 + 0.08333333333333333 * b20 - 0.02777777777777778 * b21 - 0.01388888888888889 * b22 + 0.003472222222222222 * b24 + 0.006944444444444444 * b25 - 0.02083333333333333 * b26 + 0.006944444444444444 * b27 + 0.003472222222222222 * b28 + 0.006944444444444444 * b3 + 0.003472222222222222 * b4 - 0.01388888888888889 * b6 - 0.02777777777777778 * b7 + 0.08333333333333333 * b8 - 0.02777777777777778 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 27] = 0.006944444444444444 * b1 - 0.003472222222222222 * b0 - 0.01388888888888889 * b10 - 0.02083333333333333 * b12 + 0.04166666666666667 * b13 - 0.04166666666666667 * b15 + 0.02083333333333333 * b16 + 0.01388888888888889 * b18 - 0.02777777777777778 * b19 + 0.02777777777777778 * b21 - 0.01388888888888889 * b22 - 0.003472222222222222 * b24 + 0.006944444444444444 * b25 - 0.006944444444444444 * b27 + 0.003472222222222222 * b28 - 0.006944444444444444 * b3 + 0.003472222222222222 * b4 + 0.01388888888888889 * b6 - 0.02777777777777778 * b7 + 0.02777777777777778 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 28] = 0.001736111111111111 * b0 - 0.006944444444444444 * b1 - 0.006944444444444444 * b10 + 0.01041666666666667 * b12 - 0.04166666666666667 * b13 + 0.0625 * b14 - 0.04166666666666667 * b15 + 0.01041666666666667 * b16 - 0.006944444444444444 * b18 + 0.02777777777777778 * b19 + 0.01041666666666667 * b2 - 0.04166666666666667 * b20 + 0.02777777777777778 * b21 - 0.006944444444444444 * b22 + 0.001736111111111111 * b24 - 0.006944444444444444 * b25 + 0.01041666666666667 * b26 - 0.006944444444444444 * b27 + 0.001736111111111111 * b28 - 0.006944444444444444 * b3 + 0.001736111111111111 * b4 - 0.006944444444444444 * b6 + 0.02777777777777778 * b7 - 0.04166666666666667 * b8 + 0.02777777777777778 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 29] = 0.001736111111111111 * b1 - 0.0003472222222222222 * b0 + 0.006944444444444444 * b10 - 0.001388888888888889 * b11 - 0.002083333333333333 * b12 + 0.01041666666666667 * b13 - 0.02083333333333333 * b14 + 0.02083333333333333 * b15 - 0.01041666666666667 * b16 + 0.002083333333333333 * b17 + 0.001388888888888889 * b18 - 0.006944444444444444 * b19 - 0.003472222222222222 * b2 + 0.01388888888888889 * b20 - 0.01388888888888889 * b21 + 0.006944444444444444 * b22 - 0.001388888888888889 * b23 - 0.0003472222222222222 * b24 + 0.001736111111111111 * b25 - 0.003472222222222222 * b26 + 0.003472222222222222 * b27 - 0.001736111111111111 * b28 + 0.0003472222222222222 * b29 + 0.003472222222222222 * b3 - 0.001736111111111111 * b4 + 0.0003472222222222222 * b5 + 0.001388888888888889 * b6 - 0.006944444444444444 * b7 + 0.01388888888888889 * b8 - 0.01388888888888889 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 30] = 0.0003472222222222222 * b10 - 0.001805555555555556 * b1 - 0.00006944444444444444 * b0 - 0.0006944444444444444 * b12 - 0.01805555555555556 * b13 - 0.04583333333333333 * b14 - 0.01805555555555556 * b15 - 0.0006944444444444444 * b16 + 0.0006944444444444444 * b18 + 0.01805555555555556 * b19 - 0.004583333333333333 * b2 + 0.04583333333333333 * b20 + 0.01805555555555556 * b21 + 0.0006944444444444444 * b22 - 0.0003472222222222222 * b24 - 0.009027777777777778 * b25 - 0.02291666666666667 * b26 - 0.009027777777777778 * b27 - 0.0003472222222222222 * b28 - 0.001805555555555556 * b3 + 0.00006944444444444444 * b30 + 0.001805555555555556 * b31 + 0.004583333333333333 * b32 + 0.001805555555555556 * b33 + 0.00006944444444444444 * b34 - 0.00006944444444444444 * b4 + 0.0003472222222222222 * b6 + 0.009027777777777778 * b7 + 0.02291666666666667 * b8 + 0.009027777777777778 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 31] = 0.0003472222222222222 * b0 + 0.003472222222222222 * b1 + 0.001736111111111111 * b10 + 0.003472222222222222 * b12 + 0.03472222222222222 * b13 - 0.03472222222222222 * b15 - 0.003472222222222222 * b16 - 0.003472222222222222 * b18 - 0.03472222222222222 * b19 + 0.03472222222222222 * b21 + 0.003472222222222222 * b22 + 0.001736111111111111 * b24 + 0.01736111111111111 * b25 - 0.01736111111111111 * b27 - 0.001736111111111111 * b28 - 0.003472222222222222 * b3 - 0.0003472222222222222 * b30 - 0.003472222222222222 * b31 + 0.003472222222222222 * b33 + 0.0003472222222222222 * b34 - 0.0003472222222222222 * b4 - 0.001736111111111111 * b6 - 0.01736111111111111 * b7 + 0.01736111111111111 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 32] = 0.003472222222222222 * b10 - 0.001388888888888889 * b1 - 0.0006944444444444444 * b0 - 0.006944444444444444 * b12 - 0.01388888888888889 * b13 + 0.04166666666666667 * b14 - 0.01388888888888889 * b15 - 0.006944444444444444 * b16 + 0.006944444444444444 * b18 + 0.01388888888888889 * b19 + 0.004166666666666667 * b2 - 0.04166666666666667 * b20 + 0.01388888888888889 * b21 + 0.006944444444444444 * b22 - 0.003472222222222222 * b24 - 0.006944444444444444 * b25 + 0.02083333333333333 * b26 - 0.006944444444444444 * b27 - 0.003472222222222222 * b28 - 0.001388888888888889 * b3 + 0.0006944444444444444 * b30 + 0.001388888888888889 * b31 - 0.004166666666666667 * b32 + 0.001388888888888889 * b33 + 0.0006944444444444444 * b34 - 0.0006944444444444444 * b4 + 0.003472222222222222 * b6 + 0.006944444444444444 * b7 - 0.02083333333333333 * b8 + 0.006944444444444444 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 33] = 0.0006944444444444444 * b0 - 0.001388888888888889 * b1 + 0.003472222222222222 * b10 + 0.006944444444444444 * b12 - 0.01388888888888889 * b13 + 0.01388888888888889 * b15 - 0.006944444444444444 * b16 - 0.006944444444444444 * b18 + 0.01388888888888889 * b19 - 0.01388888888888889 * b21 + 0.006944444444444444 * b22 + 0.003472222222222222 * b24 - 0.006944444444444444 * b25 + 0.006944444444444444 * b27 - 0.003472222222222222 * b28 + 0.001388888888888889 * b3 - 0.0006944444444444444 * b30 + 0.001388888888888889 * b31 - 0.001388888888888889 * b33 + 0.0006944444444444444 * b34 - 0.0006944444444444444 * b4 - 0.003472222222222222 * b6 + 0.006944444444444444 * b7 - 0.006944444444444444 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 34] = 0.001388888888888889 * b1 - 0.0003472222222222222 * b0 + 0.001736111111111111 * b10 - 0.003472222222222222 * b12 + 0.01388888888888889 * b13 - 0.02083333333333333 * b14 + 0.01388888888888889 * b15 - 0.003472222222222222 * b16 + 0.003472222222222222 * b18 - 0.01388888888888889 * b19 - 0.002083333333333333 * b2 + 0.02083333333333333 * b20 - 0.01388888888888889 * b21 + 0.003472222222222222 * b22 - 0.001736111111111111 * b24 + 0.006944444444444444 * b25 - 0.01041666666666667 * b26 + 0.006944444444444444 * b27 - 0.001736111111111111 * b28 + 0.001388888888888889 * b3 + 0.0003472222222222222 * b30 - 0.001388888888888889 * b31 + 0.002083333333333333 * b32 - 0.001388888888888889 * b33 + 0.0003472222222222222 * b34 - 0.0003472222222222222 * b4 + 0.001736111111111111 * b6 - 0.006944444444444444 * b7 + 0.01041666666666667 * b8 - 0.006944444444444444 * b9;
QK_B_QKT_buffer[lind_qkbqkt + 35] = 0.00006944444444444444 * b0 - 0.0003472222222222222 * b1 - 0.001736111111111111 * b10 + 0.0003472222222222222 * b11 + 0.0006944444444444444 * b12 - 0.003472222222222222 * b13 + 0.006944444444444444 * b14 - 0.006944444444444444 * b15 + 0.003472222222222222 * b16 - 0.0006944444444444444 * b17 - 0.0006944444444444444 * b18 + 0.003472222222222222 * b19 + 0.0006944444444444444 * b2 - 0.006944444444444444 * b20 + 0.006944444444444444 * b21 - 0.003472222222222222 * b22 + 0.0006944444444444444 * b23 + 0.0003472222222222222 * b24 - 0.001736111111111111 * b25 + 0.003472222222222222 * b26 - 0.003472222222222222 * b27 + 0.001736111111111111 * b28 - 0.0003472222222222222 * b29 - 0.0006944444444444444 * b3 - 0.00006944444444444444 * b30 + 0.0003472222222222222 * b31 - 0.0006944444444444444 * b32 + 0.0006944444444444444 * b33 - 0.0003472222222222222 * b34 + 0.00006944444444444444 * b35 + 0.0003472222222222222 * b4 - 0.00006944444444444444 * b5 - 0.0003472222222222222 * b6 + 0.001736111111111111 * b7 - 0.003472222222222222 * b8 + 0.003472222222222222 * b9;
}
}
}
void Dic::analysis(std::ofstream &of, std::size_t currImg) {
// of << "Inside analysis for current image " << currImg << "\n";
int spacing = params.subsetSpacing;
// of << "Total regions are " << roi.region.size() << "\n";
for (std::size_t num_region = 0; num_region < roi.region.size(); num_region++) {
// of << "Updating cirroi for region " << num_region << "\n";
roi.update_cirroi(num_region);
// of << "Initialize queue\n";
heap queue;
// of << "Creating file for separate qorder list\n";
// std::ofstream qf(QString("./qorder%1.txt").arg(currImg).toStdString());
// Add seed to queue
// [x y u v du/dx du/dy dv/dx dv/dy corrcoef]
std::vector<double> paramvector_seed(9, 0);
// seed is to be used here
paramvector_seed[0] = seed_info[currImg][0];
paramvector_seed[1] = seed_info[currImg][1];
paramvector_seed[2] = seed_info[currImg][2];
paramvector_seed[3] = seed_info[currImg][3];
paramvector_seed[4] = seed_info[currImg][4];
paramvector_seed[5] = seed_info[currImg][5];
paramvector_seed[6] = seed_info[currImg][6];
paramvector_seed[7] = seed_info[currImg][7];
paramvector_seed[8] = seed_info[currImg][8];
queue.push(paramvector_seed);
// of << "Seed added to queue with values : ";
// for (const auto &e : paramvector_seed)
// of << e << " ";
// of << "\n";
// of << "Inactivate seed point and mark it as calculated and valid\n";
int x_seed_reduced = static_cast<int>(paramvector_seed[0]) / (spacing + 1); // x_seed and y_seed are guaranteed divisible by (spacing+1)
int y_seed_reduced = static_cast<int>(paramvector_seed[1]) / (spacing + 1);
plot_calcpoints.value[y_seed_reduced + x_seed_reduced * plot_calcpoints.height] = true;
plot_validpoints[currImg].value[y_seed_reduced + x_seed_reduced * plot_validpoints[currImg].height] = true;
// of << "Starting while loop for queue\n";
// Enter While Loop - Exit when queue is empty
while (!queue.empty()) {
// 1) Load point with lowest correlation coefficient from queue
// 2) Delete point from queue
// 3) Add data to plots
// 4) Analyze four surrounding points and sort
// Step 1: load
std::vector<double> paramvector_init = queue.top();
//of << "Queue pop at : ";
// for (const auto &e : paramvector_init) {
//of << e << " ";
// qf << e << " ";
// }
//of << "\n";
// Step 2: delete
queue.pop();
of << paramvector_init[0] << ", " << paramvector_init[1] << "\n";
// Step 3: add data to plots
int x_init_reduced = static_cast<int>(paramvector_init[0]) / (spacing + 1);
int y_init_reduced = static_cast<int>(paramvector_init[1]) / (spacing + 1);
plot_u[currImg].value[y_init_reduced + x_init_reduced * plot_u[currImg].height] = paramvector_init[2];
plot_v[currImg].value[y_init_reduced + x_init_reduced * plot_v[currImg].height] = paramvector_init[3];
plot_corrcoef[currImg].value[y_init_reduced + x_init_reduced * plot_corrcoef[currImg].height] = paramvector_init[8];
// of << "Analysing surroinding points of (" << paramvector_init[0] << ", " << paramvector_init[1] << ")\n";
// Step 4: analyze four surrounding points - must increment by spacing parameter
analyzepoint(of, currImg, queue, static_cast<int>(paramvector_init[0]), static_cast<int>(paramvector_init[1]) - (spacing + 1), paramvector_init, num_region);
analyzepoint(of, currImg, queue, static_cast<int>(paramvector_init[0]) + (spacing + 1), static_cast<int>(paramvector_init[1]), paramvector_init, num_region);
analyzepoint(of, currImg, queue, static_cast<int>(paramvector_init[0]), static_cast<int>(paramvector_init[1]) + (spacing + 1), paramvector_init, num_region);
analyzepoint(of, currImg, queue, static_cast<int>(paramvector_init[0]) - (spacing + 1), static_cast<int>(paramvector_init[1]), paramvector_init, num_region);
}
}
}
void Dic::analyzepoint(std::ofstream &of, std::size_t currImg, heap &queue, const int &x, const int &y, const std::vector<double> ¶mvector_init, const int &num_region) {
int spacing = params.subsetSpacing;
// These are read only, so they are thread safe
static double cutoff_corrcoef = 2.0; // Heuristic, but 2.0 is pretty high. Range is [0,4]. Different images can have different corrcoef cutoffs which work well, so set this to a low value
static double cutoff_disp = spacing + 1; // Heuristic, this prevents large displacement jumps (most likely incorrect data) form being added
// Reduce coordinates first
int x_reduced = x / (spacing + 1);
int y_reduced = y / (spacing + 1);
//of << "Analysing point (" << x << ", " << y << "), reduced version (" << x_reduced << ", " << y_reduced << ")\n";
// Make sure point is within region bounds first
if (x >= roi.region[num_region].leftbound &&
x <= roi.region[num_region].rightbound &&
y >= roi.region[num_region].upperbound &&
y <= roi.region[num_region].lowerbound &&
!plot_calcpoints.value[y_reduced + x_reduced * plot_calcpoints.height] &&
roi.withinregion(x, y, num_region)) {
//of << "Point was within region bounds. Ok to proceed\n";
// Initialize paramvector
std::vector<double> paramvector(9, 0); // [x y u v du/dx du/dy dv/dx dv/dy corrcoef]
//of << "Calling calpoint for this point\n";
// Calculate paramvector for a point
bool outstate = calcpoint(of, currImg, paramvector, x, y, paramvector_init, num_region);
// Make sure parameters are correct before adding them to queue
if (outstate &&
paramvector[8] < cutoff_corrcoef &&
fabs(paramvector_init[2] - paramvector[2]) < cutoff_disp &&
fabs(paramvector_init[3] - paramvector[3]) < cutoff_disp) {
// Insert paramvector based on correlation coefficient
// of << "Calcpoint was success!\n";
queue.push(paramvector);
// Valid Point
plot_validpoints[currImg].value[y_reduced + x_reduced * plot_validpoints[currImg].height] = true;
} else {
// of << "Calcpoint failed.\n";
}
// Calculated point
plot_calcpoints.value[y_reduced + x_reduced * plot_calcpoints.height] = true;
} else {
// of << "Point not in region. Wrong point.\n";
}
}
bool Dic::calcpoint(std::ofstream &of, std::size_t currImg, std::vector<double> ¶mvector, const int &x, const int &y, const std::vector<double> ¶mvector_init, const int &num_region) {
// Get cirroi -> Find initial guess -> Refine results with IC-GN -> Return true or false and store output
// Step 1: Get cirroi
//subsettrunc = false
//of << "Called get cirroi for (" << x << ", " << y << ") point\n";
roi.get_cirroi(x, y, num_region, false);
// Step 2: Get initial guess - Use displacement and displacement gradients to get initial guess
std::vector<double> defvector_init(6, 0); // [u v du/dx du/dy dv/dx dv/dy]
// u_init = u+du/dx*x_delta+du/dy*y_delta;
// v_init = v+dv/dx*x_delta+dv/dy*y_delta;
defvector_init[0] = paramvector_init[2] + paramvector_init[4] * (x - paramvector_init[0]) + paramvector_init[5] * (y - paramvector_init[1]);
defvector_init[1] = paramvector_init[3] + paramvector_init[6] * (x - paramvector_init[0]) + paramvector_init[7] * (y - paramvector_init[1]);
defvector_init[2] = paramvector_init[4];
defvector_init[3] = paramvector_init[5];
defvector_init[4] = paramvector_init[6];
defvector_init[5] = paramvector_init[7];
// Step 3: Get refined results with IC-GN
std::vector<double> defvector(6, 0); // [u v du/dx du/dy dv/dx dv/dy]
double corrcoef;
//of << "Getting corr using iterative search\n";
bool outstate_iterative = iterativesearch(of, currImg, defvector, corrcoef, defvector_init);
if (outstate_iterative) {
// Step 4: Store output and return true
paramvector[0] = x;
paramvector[1] = y;
paramvector[2] = defvector[0];
paramvector[3] = defvector[1];
paramvector[4] = defvector[2];
paramvector[5] = defvector[3];
paramvector[6] = defvector[4];
paramvector[7] = defvector[5];
paramvector[8] = corrcoef;
//of << "Iterative search pass and calcpoint success\n";
//of << "Corr was " << corrcoef << "\n";
return true;
}
//of << "Iterative search failed!\n";
return false;
}
bool Dic::iterativesearch(std::ofstream &of, std::size_t currImg, std::vector<double> &defvector, double &corrcoef, const std::vector<double> &defvector_init) {
// Calculate fm
double fm = 0.0;
for (int i = 0; i < roi.cirroi.region.noderange.height; i++) {
for (int j = 0; j < roi.cirroi.region.noderange.value[i]; j += 2) {
for (int k = roi.cirroi.region.nodelist.value[i + j * roi.cirroi.region.nodelist.height]; k <= roi.cirroi.region.nodelist.value[i + (j + 1) * roi.cirroi.region.nodelist.height]; k++) {
int lind_ref = k + (i + (roi.cirroi.x - roi.cirroi.radius)) * rimg.gs.height;
fm += rimg.gs.value[lind_ref];
}
}
}
fm = fm / (double)roi.cirroi.region.totalpoints;
//of << "fm is " << fm << "\n";
// Calculate deltaf_inf
double deltaf_inv = 0.0;
for (int i = 0; i < roi.cirroi.region.noderange.height; i++) {
for (int j = 0; j < roi.cirroi.region.noderange.value[i]; j += 2) {
for (int k = roi.cirroi.region.nodelist.value[i + j * roi.cirroi.region.nodelist.height]; k <= roi.cirroi.region.nodelist.value[i + (j + 1) * roi.cirroi.region.nodelist.height]; k++) {
int lind_ref = k + (i + (roi.cirroi.x - roi.cirroi.radius)) * rimg.gs.height;
deltaf_inv += pow(rimg.gs.value[lind_ref] - fm, 2);
}
}
}
deltaf_inv = sqrt(deltaf_inv);
//of << "delta inf " << deltaf_inv << "\n";
// check to make sure deltaf_inv (strictly positive) isn't close to zero; if it is, iterative search fails
if (deltaf_inv > LAMBDA) {
// Finish deltaf_inv
deltaf_inv = 1.0 / deltaf_inv;
// Precompute "Steepest descent images"
for (int i = 0; i < roi.cirroi.region.noderange.height; i++) {
for (int j = 0; j < roi.cirroi.region.noderange.value[i]; j += 2) {
for (int k = roi.cirroi.region.nodelist.value[i + j * roi.cirroi.region.nodelist.height]; k <= roi.cirroi.region.nodelist.value[i + (j + 1) * roi.cirroi.region.nodelist.height]; k++) {
// Find new coordinates
double dx = (double)(i - roi.cirroi.radius);
double dy = (double)(k - roi.cirroi.y);
int y_tilda_floor = k;
int x_tilda_floor = i + (roi.cirroi.x - roi.cirroi.radius);
// Calculate lind_f for gradient and lind_df for the first order "steepest descent images"
int lind_f = y_tilda_floor + x_tilda_floor * rimg.gs.height;
int lind_df = ((k - roi.cirroi.y) + roi.cirroi.radius) * 6 + i * (roi.cirroi.region.nodelist.height * 6);
// First order
df_dp_buffer[lind_df] = df_dx_buffer[lind_f]; // u
df_dp_buffer[lind_df + 1] = df_dy_buffer[lind_f]; // v
df_dp_buffer[lind_df + 2] = df_dx_buffer[lind_f] * dx; // dudx
df_dp_buffer[lind_df + 3] = df_dx_buffer[lind_f] * dy; // dudy
df_dp_buffer[lind_df + 4] = df_dy_buffer[lind_f] * dx; // dvdx
df_dp_buffer[lind_df + 5] = df_dy_buffer[lind_f] * dy; // dvdy
}
}
}
// Precompute GN hessian
// Initialize to zero first
std::fill(hessian_gn_buffer.begin(), hessian_gn_buffer.end(), 0.0);
for (int i = 0; i < roi.cirroi.region.noderange.height; i++) {
for (int j = 0; j < roi.cirroi.region.noderange.value[i]; j += 2) {
for (int k = roi.cirroi.region.nodelist.value[i + j * roi.cirroi.region.nodelist.height]; k <= roi.cirroi.region.nodelist.value[i + (j + 1) * roi.cirroi.region.nodelist.height]; k++) {
// Parameters
int lind_df = ((k - roi.cirroi.y) + roi.cirroi.radius) * 6 + i * (roi.cirroi.region.nodelist.height * 6);
// Hessian - only calculate lower half since hessian is symmetric
hessian_gn_buffer[0] += df_dp_buffer[lind_df] * df_dp_buffer[lind_df];
hessian_gn_buffer[1] += df_dp_buffer[lind_df] * df_dp_buffer[lind_df + 1];
hessian_gn_buffer[2] += df_dp_buffer[lind_df] * df_dp_buffer[lind_df + 2];
hessian_gn_buffer[3] += df_dp_buffer[lind_df] * df_dp_buffer[lind_df + 3];
hessian_gn_buffer[4] += df_dp_buffer[lind_df] * df_dp_buffer[lind_df + 4];
hessian_gn_buffer[5] += df_dp_buffer[lind_df] * df_dp_buffer[lind_df + 5];
hessian_gn_buffer[7] += df_dp_buffer[lind_df + 1] * df_dp_buffer[lind_df + 1];
hessian_gn_buffer[8] += df_dp_buffer[lind_df + 1] * df_dp_buffer[lind_df + 2];
hessian_gn_buffer[9] += df_dp_buffer[lind_df + 1] * df_dp_buffer[lind_df + 3];
hessian_gn_buffer[10] += df_dp_buffer[lind_df + 1] * df_dp_buffer[lind_df + 4];
hessian_gn_buffer[11] += df_dp_buffer[lind_df + 1] * df_dp_buffer[lind_df + 5];
hessian_gn_buffer[14] += df_dp_buffer[lind_df + 2] * df_dp_buffer[lind_df + 2];
hessian_gn_buffer[15] += df_dp_buffer[lind_df + 2] * df_dp_buffer[lind_df + 3];
hessian_gn_buffer[16] += df_dp_buffer[lind_df + 2] * df_dp_buffer[lind_df + 4];
hessian_gn_buffer[17] += df_dp_buffer[lind_df + 2] * df_dp_buffer[lind_df + 5];
hessian_gn_buffer[21] += df_dp_buffer[lind_df + 3] * df_dp_buffer[lind_df + 3];
hessian_gn_buffer[22] += df_dp_buffer[lind_df + 3] * df_dp_buffer[lind_df + 4];
hessian_gn_buffer[23] += df_dp_buffer[lind_df + 3] * df_dp_buffer[lind_df + 5];
hessian_gn_buffer[28] += df_dp_buffer[lind_df + 4] * df_dp_buffer[lind_df + 4];
hessian_gn_buffer[29] += df_dp_buffer[lind_df + 4] * df_dp_buffer[lind_df + 5];
hessian_gn_buffer[35] += df_dp_buffer[lind_df + 5] * df_dp_buffer[lind_df + 5];
}
}
}
// Multiply components of hessian by 2/deltaf^2
for (int i = 0; i < 6; i++) {
for (int j = i; j < 6; j++) {
hessian_gn_buffer[j + i * 6] *= 2 * pow(deltaf_inv, 2);
}
}
// Fill other half of hessian
hessian_gn_buffer[6] = hessian_gn_buffer[1];
hessian_gn_buffer[12] = hessian_gn_buffer[2];
hessian_gn_buffer[13] = hessian_gn_buffer[8];
hessian_gn_buffer[18] = hessian_gn_buffer[3];
hessian_gn_buffer[19] = hessian_gn_buffer[9];
hessian_gn_buffer[20] = hessian_gn_buffer[15];
hessian_gn_buffer[24] = hessian_gn_buffer[4];
hessian_gn_buffer[25] = hessian_gn_buffer[10];
hessian_gn_buffer[26] = hessian_gn_buffer[16];
hessian_gn_buffer[27] = hessian_gn_buffer[22];
hessian_gn_buffer[30] = hessian_gn_buffer[5];
hessian_gn_buffer[31] = hessian_gn_buffer[11];
hessian_gn_buffer[32] = hessian_gn_buffer[17];
hessian_gn_buffer[33] = hessian_gn_buffer[23];
hessian_gn_buffer[34] = hessian_gn_buffer[29];
// Solve for new parameters via cholesky decomp (from Golub and Van Loan)
// Lower triangle of Hessian overwritten with parameters used in Cholesky decomp
// If one of the diagonals is close to zero or negative, then the
// hessian is not positive definite
bool positivedef = true;
dicutils::cholesky(hessian_gn_buffer, positivedef, 6);
if (positivedef) {
// Start iterations - For first iteration use defvector_init
double diffnorm;
bool outstate_newton = newton(of, currImg, defvector, corrcoef, diffnorm, defvector_init, fm, deltaf_inv);
// Initialize counter
int counter = 1;
while (outstate_newton && diffnorm >= params.cutoff_diffnorm && counter <= params.cutoff_iteration) {
// For rest of iterations use defvector from previous iterations
outstate_newton = newton(of, currImg, defvector, corrcoef, diffnorm, defvector, fm, deltaf_inv);
++counter;
}
if (outstate_newton) {
return true;
}
}
}
// Some parameters are invalid - either deltag_inv was zero or the hessian wasn't positive definite
return false;
}
bool Dic::newton(std::ofstream &of, std::size_t currImg, std::vector<double> &defvector, double &corrcoef, double &diffnorm, const std::vector<double> &defvector_init, const double &fm, const double &deltaf_inv) {
// Will only overwrite queue_new if parameters are valid
// Interpolate g subset - do this here instead of interp_qbs because QK_B_QKT has been precomputed
double gm = 0.0;
for (int i = 0; i < roi.cirroi.region.noderange.height; i++) {
for (int j = 0; j < roi.cirroi.region.noderange.value[i]; j += 2) {
for (int k = roi.cirroi.region.nodelist.value[i + j * roi.cirroi.region.nodelist.height]; k <= roi.cirroi.region.nodelist.value[i + (j + 1) * roi.cirroi.region.nodelist.height]; k++) {
// Find new coordinates
double dx = (double)(i - roi.cirroi.radius);
double dy = (double)(k - roi.cirroi.y);
double y_tilda = (double)k + defvector_init[1] + defvector_init[4] * dx + defvector_init[5] * dy;
double x_tilda = (double)(i + (roi.cirroi.x - roi.cirroi.radius)) + defvector_init[0] + defvector_init[2] * dx + defvector_init[3] * dy;
int x_tilda_floor = (int)floor(x_tilda);
int y_tilda_floor = (int)floor(y_tilda);
int lind_g = (int)dy + roi.cirroi.radius + i * roi.cirroi.region.nodelist.height;
// Get bounds of the desired b-spline coefficients used for interpolation
int top = y_tilda_floor + cimgs[currImg].border_bcoef - 2;
int left = x_tilda_floor + cimgs[currImg].border_bcoef - 2;
int bottom = y_tilda_floor + cimgs[currImg].border_bcoef + 3;
int right = x_tilda_floor + cimgs[currImg].border_bcoef + 3;
if (top >= 0 &&
left >= 0 &&
bottom < cimgs[currImg].bcoef.height &&
right < cimgs[currImg].bcoef.width) {
double x_tilda_delta = x_tilda - (double)x_tilda_floor;
double y_tilda_delta = y_tilda - (double)y_tilda_floor;
// Form x_vec
x_vec_buffer[1] = x_tilda_delta;
x_vec_buffer[2] = x_tilda_delta * x_tilda_delta;
x_vec_buffer[3] = x_tilda_delta * x_tilda_delta * x_tilda_delta;
x_vec_buffer[4] = x_tilda_delta * x_tilda_delta * x_tilda_delta * x_tilda_delta;
x_vec_buffer[5] = x_tilda_delta * x_tilda_delta * x_tilda_delta * x_tilda_delta * x_tilda_delta;
// Form y_vec
y_vec_buffer[1] = y_tilda_delta;
y_vec_buffer[2] = y_tilda_delta * y_tilda_delta;
y_vec_buffer[3] = y_tilda_delta * y_tilda_delta * y_tilda_delta;
y_vec_buffer[4] = y_tilda_delta * y_tilda_delta * y_tilda_delta * y_tilda_delta;
y_vec_buffer[5] = y_tilda_delta * y_tilda_delta * y_tilda_delta * y_tilda_delta * y_tilda_delta;
// Calculate lind_qkbqkt for QK_B_QKT
int lind_qkbqkt = (top * 36) + (left * 36 * (cimgs[currImg].bcoef.height - 5));
// Get QK_B_QKT coefficients
double QK_B_QKT_0 = QK_B_QKT_buffer[lind_qkbqkt];
double QK_B_QKT_1 = QK_B_QKT_buffer[lind_qkbqkt + 1];
double QK_B_QKT_2 = QK_B_QKT_buffer[lind_qkbqkt + 2];
double QK_B_QKT_3 = QK_B_QKT_buffer[lind_qkbqkt + 3];
double QK_B_QKT_4 = QK_B_QKT_buffer[lind_qkbqkt + 4];
double QK_B_QKT_5 = QK_B_QKT_buffer[lind_qkbqkt + 5];
double QK_B_QKT_6 = QK_B_QKT_buffer[lind_qkbqkt + 6];
double QK_B_QKT_7 = QK_B_QKT_buffer[lind_qkbqkt + 7];
double QK_B_QKT_8 = QK_B_QKT_buffer[lind_qkbqkt + 8];
double QK_B_QKT_9 = QK_B_QKT_buffer[lind_qkbqkt + 9];
double QK_B_QKT_10 = QK_B_QKT_buffer[lind_qkbqkt + 10];
double QK_B_QKT_11 = QK_B_QKT_buffer[lind_qkbqkt + 11];
double QK_B_QKT_12 = QK_B_QKT_buffer[lind_qkbqkt + 12];
double QK_B_QKT_13 = QK_B_QKT_buffer[lind_qkbqkt + 13];
double QK_B_QKT_14 = QK_B_QKT_buffer[lind_qkbqkt + 14];
double QK_B_QKT_15 = QK_B_QKT_buffer[lind_qkbqkt + 15];
double QK_B_QKT_16 = QK_B_QKT_buffer[lind_qkbqkt + 16];
double QK_B_QKT_17 = QK_B_QKT_buffer[lind_qkbqkt + 17];
double QK_B_QKT_18 = QK_B_QKT_buffer[lind_qkbqkt + 18];
double QK_B_QKT_19 = QK_B_QKT_buffer[lind_qkbqkt + 19];
double QK_B_QKT_20 = QK_B_QKT_buffer[lind_qkbqkt + 20];
double QK_B_QKT_21 = QK_B_QKT_buffer[lind_qkbqkt + 21];
double QK_B_QKT_22 = QK_B_QKT_buffer[lind_qkbqkt + 22];
double QK_B_QKT_23 = QK_B_QKT_buffer[lind_qkbqkt + 23];
double QK_B_QKT_24 = QK_B_QKT_buffer[lind_qkbqkt + 24];
double QK_B_QKT_25 = QK_B_QKT_buffer[lind_qkbqkt + 25];
double QK_B_QKT_26 = QK_B_QKT_buffer[lind_qkbqkt + 26];
double QK_B_QKT_27 = QK_B_QKT_buffer[lind_qkbqkt + 27];
double QK_B_QKT_28 = QK_B_QKT_buffer[lind_qkbqkt + 28];
double QK_B_QKT_29 = QK_B_QKT_buffer[lind_qkbqkt + 29];
double QK_B_QKT_30 = QK_B_QKT_buffer[lind_qkbqkt + 30];
double QK_B_QKT_31 = QK_B_QKT_buffer[lind_qkbqkt + 31];
double QK_B_QKT_32 = QK_B_QKT_buffer[lind_qkbqkt + 32];
double QK_B_QKT_33 = QK_B_QKT_buffer[lind_qkbqkt + 33];
double QK_B_QKT_34 = QK_B_QKT_buffer[lind_qkbqkt + 34];
double QK_B_QKT_35 = QK_B_QKT_buffer[lind_qkbqkt + 35];
// Calculate g - main computational bottleneck of the inverse compositional method with biquintic b-splines
g_buffer[lind_g] = (QK_B_QKT_0 + x_vec_buffer[1] * QK_B_QKT_6 + x_vec_buffer[2] * QK_B_QKT_12 + x_vec_buffer[3] * QK_B_QKT_18 + x_vec_buffer[4] * QK_B_QKT_24 + x_vec_buffer[5] * QK_B_QKT_30) +
(QK_B_QKT_1 + x_vec_buffer[1] * QK_B_QKT_7 + x_vec_buffer[2] * QK_B_QKT_13 + x_vec_buffer[3] * QK_B_QKT_19 + x_vec_buffer[4] * QK_B_QKT_25 + x_vec_buffer[5] * QK_B_QKT_31) * y_vec_buffer[1] +
(QK_B_QKT_2 + x_vec_buffer[1] * QK_B_QKT_8 + x_vec_buffer[2] * QK_B_QKT_14 + x_vec_buffer[3] * QK_B_QKT_20 + x_vec_buffer[4] * QK_B_QKT_26 + x_vec_buffer[5] * QK_B_QKT_32) * y_vec_buffer[2] +
(QK_B_QKT_3 + x_vec_buffer[1] * QK_B_QKT_9 + x_vec_buffer[2] * QK_B_QKT_15 + x_vec_buffer[3] * QK_B_QKT_21 + x_vec_buffer[4] * QK_B_QKT_27 + x_vec_buffer[5] * QK_B_QKT_33) * y_vec_buffer[3] +
(QK_B_QKT_4 + x_vec_buffer[1] * QK_B_QKT_10 + x_vec_buffer[2] * QK_B_QKT_16 + x_vec_buffer[3] * QK_B_QKT_22 + x_vec_buffer[4] * QK_B_QKT_28 + x_vec_buffer[5] * QK_B_QKT_34) * y_vec_buffer[4] +
(QK_B_QKT_5 + x_vec_buffer[1] * QK_B_QKT_11 + x_vec_buffer[2] * QK_B_QKT_17 + x_vec_buffer[3] * QK_B_QKT_23 + x_vec_buffer[4] * QK_B_QKT_29 + x_vec_buffer[5] * QK_B_QKT_35) * y_vec_buffer[5];
// Add components to calculate the mean
gm += g_buffer[lind_g];
} else {
// If this condition is satisfied then we are
// interpolating a point beyond the bounds of the
// original image, so just set the values to zero
g_buffer[lind_g] = 0.0;
// Don't add anything to averages
continue;
}
}
}
}
// Divide by totalpoints to get real average
gm /= static_cast<double>(roi.cirroi.region.totalpoints);
// Calculate deltag_inv
double deltag_inv = 0.0;
for (int i = 0; i < roi.cirroi.region.noderange.height; i++) {
for (int j = 0; j < roi.cirroi.region.noderange.value[i]; j += 2) {
for (int k = roi.cirroi.region.nodelist.value[i + j * roi.cirroi.region.nodelist.height]; k <= roi.cirroi.region.nodelist.value[i + (j + 1) * roi.cirroi.region.nodelist.height]; k++) {
int lind_g = (k - roi.cirroi.y) + roi.cirroi.radius + i * roi.cirroi.region.nodelist.height;
deltag_inv = deltag_inv + pow(g_buffer[lind_g] - gm, 2);
}
}
}
deltag_inv = sqrt(deltag_inv); // This is deltag; will take inverse after ensuring it is not close to zero
// check to make sure deltag_inv (strictly positive) isn't close to zero; if it is, exit newton raphson
if (deltag_inv > LAMBDA) {
// Finish deltag_inv
deltag_inv = 1.0 / deltag_inv;
// Calculate gradient
// Initialize to zero first
std::fill(gradient_buffer.begin(), gradient_buffer.end(), 0.0);
corrcoef = 0.0;
for (int i = 0; i < roi.cirroi.region.noderange.height; i++) {
for (int j = 0; j < roi.cirroi.region.noderange.value[i]; j += 2) {
for (int k = roi.cirroi.region.nodelist.value[i + j * roi.cirroi.region.nodelist.height]; k <= roi.cirroi.region.nodelist.value[i + (j + 1) * roi.cirroi.region.nodelist.height]; k++) {
// Parameters
int lind_f = k + (i + (roi.cirroi.x - roi.cirroi.radius)) * rimg.gs.height;
int lind_df = ((k - roi.cirroi.y) + roi.cirroi.radius) * 6 + i * (roi.cirroi.region.nodelist.height * 6);
int lind_g = (k - roi.cirroi.y) + roi.cirroi.radius + i * roi.cirroi.region.nodelist.height;
// Gradient Parameters
double normalized_diff = (rimg.gs.value[lind_f] - fm) * deltaf_inv - (g_buffer[lind_g] - gm) * deltag_inv;
// Gradient
gradient_buffer[0] += normalized_diff * df_dp_buffer[lind_df];
gradient_buffer[1] += normalized_diff * df_dp_buffer[lind_df + 1];
gradient_buffer[2] += normalized_diff * df_dp_buffer[lind_df + 2];
gradient_buffer[3] += normalized_diff * df_dp_buffer[lind_df + 3];
gradient_buffer[4] += normalized_diff * df_dp_buffer[lind_df + 4];
gradient_buffer[5] += normalized_diff * df_dp_buffer[lind_df + 5];
// Correlation coefficient
corrcoef += pow(normalized_diff, 2);
}
}
}
// Update gradient; multiply by the inverses
for (int i = 0; i < 6; i++) {
gradient_buffer[i] *= 2 * deltaf_inv;
}
// Find new change in deformation parameters
// Ax = b
// GG'x = b, where G is lower triangular
// Gy = b -> G'x = y
// Step 1: solve for y with forward substitution; y is stored in gradient_buffer
dicutils::forwardsub(gradient_buffer, hessian_gn_buffer, 6);
// Step 2: solve for x with back substitution
dicutils::backwardsub(gradient_buffer, hessian_gn_buffer, 6);
// Make gradient_buffer negative
for (int i = 0; i < 6; i++) {
gradient_buffer[i] = -gradient_buffer[i];
}
// At this point the change in deformation parameters is stored in gradient_buffer
// Calculate difference norm - this is stored in gradient_buffer at this point
diffnorm = 0.0;
for (int i = 0; i < 6; i++) {
diffnorm += gradient_buffer[i] * gradient_buffer[i];
}
diffnorm = sqrt(diffnorm);
// Update parameters using inverse composition
// Transfer parameters because defvector_init is an alias of defvector after the first iteration
double defvector_init_u = defvector_init[0];
double defvector_init_v = defvector_init[1];
double defvector_init_dudx = defvector_init[2];
double defvector_init_dudy = defvector_init[3];
double defvector_init_dvdx = defvector_init[4];
double defvector_init_dvdy = defvector_init[5];
defvector[0] = defvector_init_u - ((defvector_init_dudx + 1) * (gradient_buffer[0] + gradient_buffer[0] * gradient_buffer[5] - gradient_buffer[1] * gradient_buffer[3])) / (gradient_buffer[2] + gradient_buffer[5] + gradient_buffer[2] * gradient_buffer[5] - gradient_buffer[3] * gradient_buffer[4] + 1) - (defvector_init_dudy * (gradient_buffer[1] - gradient_buffer[0] * gradient_buffer[4] + gradient_buffer[1] * gradient_buffer[2])) / (gradient_buffer[2] + gradient_buffer[5] + gradient_buffer[2] * gradient_buffer[5] - gradient_buffer[3] * gradient_buffer[4] + 1); // u
defvector[1] = defvector_init_v - ((defvector_init_dvdy + 1) * (gradient_buffer[1] - gradient_buffer[0] * gradient_buffer[4] + gradient_buffer[1] * gradient_buffer[2])) / (gradient_buffer[2] + gradient_buffer[5] + gradient_buffer[2] * gradient_buffer[5] - gradient_buffer[3] * gradient_buffer[4] + 1) - (defvector_init_dvdx * (gradient_buffer[0] + gradient_buffer[0] * gradient_buffer[5] - gradient_buffer[1] * gradient_buffer[3])) / (gradient_buffer[2] + gradient_buffer[5] + gradient_buffer[2] * gradient_buffer[5] - gradient_buffer[3] * gradient_buffer[4] + 1); // v
defvector[2] = ((gradient_buffer[5] + 1) * (defvector_init_dudx + 1)) / (gradient_buffer[2] + gradient_buffer[5] + gradient_buffer[2] * gradient_buffer[5] - gradient_buffer[3] * gradient_buffer[4] + 1) - (gradient_buffer[4] * defvector_init_dudy) / (gradient_buffer[2] + gradient_buffer[5] + gradient_buffer[2] * gradient_buffer[5] - gradient_buffer[3] * gradient_buffer[4] + 1) - 1; // du/dx
defvector[3] = (defvector_init_dudy * (gradient_buffer[2] + 1)) / (gradient_buffer[2] + gradient_buffer[5] + gradient_buffer[2] * gradient_buffer[5] - gradient_buffer[3] * gradient_buffer[4] + 1) - (gradient_buffer[3] * (defvector_init_dudx + 1)) / (gradient_buffer[2] + gradient_buffer[5] + gradient_buffer[2] * gradient_buffer[5] - gradient_buffer[3] * gradient_buffer[4] + 1); // du/dy
defvector[4] = (defvector_init_dvdx * (gradient_buffer[5] + 1)) / (gradient_buffer[2] + gradient_buffer[5] + gradient_buffer[2] * gradient_buffer[5] - gradient_buffer[3] * gradient_buffer[4] + 1) - (gradient_buffer[4] * (defvector_init_dvdy + 1)) / (gradient_buffer[2] + gradient_buffer[5] + gradient_buffer[2] * gradient_buffer[5] - gradient_buffer[3] * gradient_buffer[4] + 1); // dv/dx
defvector[5] = ((gradient_buffer[2] + 1) * (defvector_init_dvdy + 1)) / (gradient_buffer[2] + gradient_buffer[5] + gradient_buffer[2] * gradient_buffer[5] - gradient_buffer[3] * gradient_buffer[4] + 1) - (gradient_buffer[3] * defvector_init_dvdx) / (gradient_buffer[2] + gradient_buffer[5] + gradient_buffer[2] * gradient_buffer[5] - gradient_buffer[3] * gradient_buffer[4] + 1) - 1; // dv/dy
// Return successful
return true;
}
// Deltag was close to zero - return failed
return false;
}
void Dic::matchSeed(std::size_t currentIndex) {
double maxCorrelation = 0.0, correlation;
std::pair<int, int> match, candidate;
std::vector<double> serialSeed =
serializeSubset(rimg, params.seedPoint);
for (int i = 0; i < cimgs[currentIndex].gs.height; i++) {
for (int j = 0; j < cimgs[currentIndex].gs.width; j++) {
candidate = std::make_pair(i, j);
correlation = dicutils::ncc(
serialSeed, serializeSubset(cimgs[currentIndex], candidate));
if (correlation > maxCorrelation) {
match = candidate;
maxCorrelation = correlation;
}
}
}
/*qDebug() << QString("for %1 th current image, found match at {%2, %3} with "
"correlation = %4")
.arg(currentIndex)
.arg(match.first)
.arg(match.second)
.arg(maxCorrelation);*/
}
std::vector<double> Dic::serializeSubset(class_img &image,
std::pair<int, int> center) {
unsigned long xMin = 0;
unsigned long xMax = static_cast<unsigned long>(image.gs.width) - 1;
unsigned long yMin = 0;
unsigned long yMax = static_cast<unsigned long>(image.gs.height) - 1;
unsigned long side = static_cast<unsigned long>(params.subsetSize);
unsigned long xStart = static_cast<unsigned long>(center.first) - side / 2;
unsigned long yStart = static_cast<unsigned long>(center.second) - side / 2;
unsigned long x;
unsigned long y;
std::vector<double> res(static_cast<size_t>(side * side), 0.0);
if (xStart < xMin or yStart < yMin or xStart + side - 1 > xMax or
yStart + side - 1 > yMax) {
std::vector<double> empty;
return empty;
}
x = xStart;
/*for (size_t i = 0; i < side; i++, x++) {
y = yStart;
for (size_t j = 0; j < side; j++, y++) {
res[i * side + j] = static_cast<double>(image.getValue(static_cast<int>(x), static_cast<int>(y)));
}
}*/
return res;
}