forked from opencv/opencv
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathtvl1flow.cpp
More file actions
1477 lines (1233 loc) · 49.1 KB
/
tvl1flow.cpp
File metadata and controls
1477 lines (1233 loc) · 49.1 KB
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
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
/*
//
// This implementation is based on Javier Sánchez Pérez <jsanchez@dis.ulpgc.es> implementation.
// Original BSD license:
//
// Copyright (c) 2011, Javier Sánchez Pérez, Enric Meinhardt Llopis
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are met:
//
// * Redistributions of source code must retain the above copyright notice, this
// list of conditions and the following disclaimer.
//
// * Redistributions in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
*/
#include "precomp.hpp"
#include "opencl_kernels_video.hpp"
#include <limits>
#include <iomanip>
#include <iostream>
#include "opencv2/core/opencl/ocl_defs.hpp"
using namespace cv;
namespace {
class OpticalFlowDual_TVL1 : public DualTVL1OpticalFlow
{
public:
OpticalFlowDual_TVL1(double tau_, double lambda_, double theta_, int nscales_, int warps_,
double epsilon_, int innerIterations_, int outerIterations_,
double scaleStep_, double gamma_, int medianFiltering_,
bool useInitialFlow_) :
tau(tau_), lambda(lambda_), theta(theta_), gamma(gamma_), nscales(nscales_),
warps(warps_), epsilon(epsilon_), innerIterations(innerIterations_),
outerIterations(outerIterations_), useInitialFlow(useInitialFlow_),
scaleStep(scaleStep_), medianFiltering(medianFiltering_)
{
}
OpticalFlowDual_TVL1();
void calc(InputArray I0, InputArray I1, InputOutputArray flow);
void collectGarbage();
CV_IMPL_PROPERTY(double, Tau, tau)
CV_IMPL_PROPERTY(double, Lambda, lambda)
CV_IMPL_PROPERTY(double, Theta, theta)
CV_IMPL_PROPERTY(double, Gamma, gamma)
CV_IMPL_PROPERTY(int, ScalesNumber, nscales)
CV_IMPL_PROPERTY(int, WarpingsNumber, warps)
CV_IMPL_PROPERTY(double, Epsilon, epsilon)
CV_IMPL_PROPERTY(int, InnerIterations, innerIterations)
CV_IMPL_PROPERTY(int, OuterIterations, outerIterations)
CV_IMPL_PROPERTY(bool, UseInitialFlow, useInitialFlow)
CV_IMPL_PROPERTY(double, ScaleStep, scaleStep)
CV_IMPL_PROPERTY(int, MedianFiltering, medianFiltering)
protected:
double tau;
double lambda;
double theta;
double gamma;
int nscales;
int warps;
double epsilon;
int innerIterations;
int outerIterations;
bool useInitialFlow;
double scaleStep;
int medianFiltering;
private:
void procOneScale(const Mat_<float>& I0, const Mat_<float>& I1, Mat_<float>& u1, Mat_<float>& u2, Mat_<float>& u3);
#ifdef HAVE_OPENCL
bool procOneScale_ocl(const UMat& I0, const UMat& I1, UMat& u1, UMat& u2);
bool calc_ocl(InputArray I0, InputArray I1, InputOutputArray flow);
#endif
struct dataMat
{
std::vector<Mat_<float> > I0s;
std::vector<Mat_<float> > I1s;
std::vector<Mat_<float> > u1s;
std::vector<Mat_<float> > u2s;
std::vector<Mat_<float> > u3s;
Mat_<float> I1x_buf;
Mat_<float> I1y_buf;
Mat_<float> flowMap1_buf;
Mat_<float> flowMap2_buf;
Mat_<float> I1w_buf;
Mat_<float> I1wx_buf;
Mat_<float> I1wy_buf;
Mat_<float> grad_buf;
Mat_<float> rho_c_buf;
Mat_<float> v1_buf;
Mat_<float> v2_buf;
Mat_<float> v3_buf;
Mat_<float> p11_buf;
Mat_<float> p12_buf;
Mat_<float> p21_buf;
Mat_<float> p22_buf;
Mat_<float> p31_buf;
Mat_<float> p32_buf;
Mat_<float> div_p1_buf;
Mat_<float> div_p2_buf;
Mat_<float> div_p3_buf;
Mat_<float> u1x_buf;
Mat_<float> u1y_buf;
Mat_<float> u2x_buf;
Mat_<float> u2y_buf;
Mat_<float> u3x_buf;
Mat_<float> u3y_buf;
} dm;
#ifdef HAVE_OPENCL
struct dataUMat
{
std::vector<UMat> I0s;
std::vector<UMat> I1s;
std::vector<UMat> u1s;
std::vector<UMat> u2s;
UMat I1x_buf;
UMat I1y_buf;
UMat I1w_buf;
UMat I1wx_buf;
UMat I1wy_buf;
UMat grad_buf;
UMat rho_c_buf;
UMat p11_buf;
UMat p12_buf;
UMat p21_buf;
UMat p22_buf;
UMat diff_buf;
UMat norm_buf;
} dum;
#endif
};
#ifdef HAVE_OPENCL
namespace cv_ocl_tvl1flow
{
bool centeredGradient(const UMat &src, UMat &dx, UMat &dy);
bool warpBackward(const UMat &I0, const UMat &I1, UMat &I1x, UMat &I1y,
UMat &u1, UMat &u2, UMat &I1w, UMat &I1wx, UMat &I1wy,
UMat &grad, UMat &rho);
bool estimateU(UMat &I1wx, UMat &I1wy, UMat &grad,
UMat &rho_c, UMat &p11, UMat &p12,
UMat &p21, UMat &p22, UMat &u1,
UMat &u2, UMat &error, float l_t, float theta, char calc_error);
bool estimateDualVariables(UMat &u1, UMat &u2,
UMat &p11, UMat &p12, UMat &p21, UMat &p22, float taut);
}
bool cv_ocl_tvl1flow::centeredGradient(const UMat &src, UMat &dx, UMat &dy)
{
size_t globalsize[2] = { (size_t)src.cols, (size_t)src.rows };
ocl::Kernel kernel;
if (!kernel.create("centeredGradientKernel", cv::ocl::video::optical_flow_tvl1_oclsrc, ""))
return false;
int idxArg = 0;
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(src));//src mat
idxArg = kernel.set(idxArg, (int)(src.cols));//src mat col
idxArg = kernel.set(idxArg, (int)(src.rows));//src mat rows
idxArg = kernel.set(idxArg, (int)(src.step / src.elemSize()));//src mat step
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrWriteOnly(dx));//res mat dx
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrWriteOnly(dy));//res mat dy
idxArg = kernel.set(idxArg, (int)(dx.step/dx.elemSize()));//res mat step
return kernel.run(2, globalsize, NULL, false);
}
bool cv_ocl_tvl1flow::warpBackward(const UMat &I0, const UMat &I1, UMat &I1x, UMat &I1y,
UMat &u1, UMat &u2, UMat &I1w, UMat &I1wx, UMat &I1wy,
UMat &grad, UMat &rho)
{
size_t globalsize[2] = { (size_t)I0.cols, (size_t)I0.rows };
ocl::Kernel kernel;
if (!kernel.create("warpBackwardKernel", cv::ocl::video::optical_flow_tvl1_oclsrc, ""))
return false;
int idxArg = 0;
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(I0));//I0 mat
int I0_step = (int)(I0.step / I0.elemSize());
idxArg = kernel.set(idxArg, I0_step);//I0_step
idxArg = kernel.set(idxArg, (int)(I0.cols));//I0_col
idxArg = kernel.set(idxArg, (int)(I0.rows));//I0_row
ocl::Image2D imageI1(I1);
ocl::Image2D imageI1x(I1x);
ocl::Image2D imageI1y(I1y);
idxArg = kernel.set(idxArg, imageI1);//image2d_t tex_I1
idxArg = kernel.set(idxArg, imageI1x);//image2d_t tex_I1x
idxArg = kernel.set(idxArg, imageI1y);//image2d_t tex_I1y
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(u1));//const float* u1
idxArg = kernel.set(idxArg, (int)(u1.step / u1.elemSize()));//int u1_step
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(u2));//const float* u2
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrWriteOnly(I1w));///float* I1w
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrWriteOnly(I1wx));//float* I1wx
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrWriteOnly(I1wy));//float* I1wy
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrWriteOnly(grad));//float* grad
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrWriteOnly(rho));//float* rho
idxArg = kernel.set(idxArg, (int)(I1w.step / I1w.elemSize()));//I1w_step
idxArg = kernel.set(idxArg, (int)(u2.step / u2.elemSize()));//u2_step
int u1_offset_x = (int)((u1.offset) % (u1.step));
u1_offset_x = (int)(u1_offset_x / u1.elemSize());
idxArg = kernel.set(idxArg, (int)u1_offset_x );//u1_offset_x
idxArg = kernel.set(idxArg, (int)(u1.offset/u1.step));//u1_offset_y
int u2_offset_x = (int)((u2.offset) % (u2.step));
u2_offset_x = (int) (u2_offset_x / u2.elemSize());
idxArg = kernel.set(idxArg, (int)u2_offset_x);//u2_offset_x
idxArg = kernel.set(idxArg, (int)(u2.offset / u2.step));//u2_offset_y
return kernel.run(2, globalsize, NULL, false);
}
bool cv_ocl_tvl1flow::estimateU(UMat &I1wx, UMat &I1wy, UMat &grad,
UMat &rho_c, UMat &p11, UMat &p12,
UMat &p21, UMat &p22, UMat &u1,
UMat &u2, UMat &error, float l_t, float theta, char calc_error)
{
size_t globalsize[2] = { (size_t)I1wx.cols, (size_t)I1wx.rows };
ocl::Kernel kernel;
if (!kernel.create("estimateUKernel", cv::ocl::video::optical_flow_tvl1_oclsrc, ""))
return false;
int idxArg = 0;
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(I1wx)); //const float* I1wx
idxArg = kernel.set(idxArg, (int)(I1wx.cols)); //int I1wx_col
idxArg = kernel.set(idxArg, (int)(I1wx.rows)); //int I1wx_row
idxArg = kernel.set(idxArg, (int)(I1wx.step/I1wx.elemSize())); //int I1wx_step
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(I1wy)); //const float* I1wy
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(grad)); //const float* grad
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(rho_c)); //const float* rho_c
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(p11)); //const float* p11
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(p12)); //const float* p12
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(p21)); //const float* p21
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(p22)); //const float* p22
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadWrite(u1)); //float* u1
idxArg = kernel.set(idxArg, (int)(u1.step / u1.elemSize())); //int u1_step
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadWrite(u2)); //float* u2
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrWriteOnly(error)); //float* error
idxArg = kernel.set(idxArg, (float)l_t); //float l_t
idxArg = kernel.set(idxArg, (float)theta); //float theta
idxArg = kernel.set(idxArg, (int)(u2.step / u2.elemSize()));//int u2_step
int u1_offset_x = (int)(u1.offset % u1.step);
u1_offset_x = (int) (u1_offset_x / u1.elemSize());
idxArg = kernel.set(idxArg, (int)u1_offset_x); //int u1_offset_x
idxArg = kernel.set(idxArg, (int)(u1.offset/u1.step)); //int u1_offset_y
int u2_offset_x = (int)(u2.offset % u2.step);
u2_offset_x = (int)(u2_offset_x / u2.elemSize());
idxArg = kernel.set(idxArg, (int)u2_offset_x ); //int u2_offset_x
idxArg = kernel.set(idxArg, (int)(u2.offset / u2.step)); //int u2_offset_y
idxArg = kernel.set(idxArg, (char)calc_error); //char calc_error
return kernel.run(2, globalsize, NULL, false);
}
bool cv_ocl_tvl1flow::estimateDualVariables(UMat &u1, UMat &u2,
UMat &p11, UMat &p12, UMat &p21, UMat &p22, float taut)
{
size_t globalsize[2] = { (size_t)u1.cols, (size_t)u1.rows };
ocl::Kernel kernel;
if (!kernel.create("estimateDualVariablesKernel", cv::ocl::video::optical_flow_tvl1_oclsrc, ""))
return false;
int idxArg = 0;
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(u1));// const float* u1
idxArg = kernel.set(idxArg, (int)(u1.cols)); //int u1_col
idxArg = kernel.set(idxArg, (int)(u1.rows)); //int u1_row
idxArg = kernel.set(idxArg, (int)(u1.step/u1.elemSize())); //int u1_step
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(u2)); // const float* u2
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadWrite(p11)); // float* p11
idxArg = kernel.set(idxArg, (int)(p11.step/p11.elemSize())); //int p11_step
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadWrite(p12)); // float* p12
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadWrite(p21)); // float* p21
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadWrite(p22)); // float* p22
idxArg = kernel.set(idxArg, (float)(taut)); //float taut
idxArg = kernel.set(idxArg, (int)(u2.step/u2.elemSize())); //int u2_step
int u1_offset_x = (int)(u1.offset % u1.step);
u1_offset_x = (int)(u1_offset_x / u1.elemSize());
idxArg = kernel.set(idxArg, u1_offset_x); //int u1_offset_x
idxArg = kernel.set(idxArg, (int)(u1.offset / u1.step)); //int u1_offset_y
int u2_offset_x = (int)(u2.offset % u2.step);
u2_offset_x = (int)(u2_offset_x / u2.elemSize());
idxArg = kernel.set(idxArg, u2_offset_x); //int u2_offset_x
idxArg = kernel.set(idxArg, (int)(u2.offset / u2.step)); //int u2_offset_y
return kernel.run(2, globalsize, NULL, false);
}
#endif
OpticalFlowDual_TVL1::OpticalFlowDual_TVL1()
{
tau = 0.25;
lambda = 0.15;
theta = 0.3;
nscales = 5;
warps = 5;
epsilon = 0.01;
gamma = 0.;
innerIterations = 30;
outerIterations = 10;
useInitialFlow = false;
medianFiltering = 5;
scaleStep = 0.8;
}
void OpticalFlowDual_TVL1::calc(InputArray _I0, InputArray _I1, InputOutputArray _flow)
{
CV_INSTRUMENT_REGION()
#ifndef __APPLE__
CV_OCL_RUN(_flow.isUMat() &&
ocl::Image2D::isFormatSupported(CV_32F, 1, false),
calc_ocl(_I0, _I1, _flow))
#endif
Mat I0 = _I0.getMat();
Mat I1 = _I1.getMat();
CV_Assert( I0.type() == CV_8UC1 || I0.type() == CV_32FC1 );
CV_Assert( I0.size() == I1.size() );
CV_Assert( I0.type() == I1.type() );
CV_Assert( !useInitialFlow || (_flow.size() == I0.size() && _flow.type() == CV_32FC2) );
CV_Assert( nscales > 0 );
bool use_gamma = gamma != 0;
// allocate memory for the pyramid structure
dm.I0s.resize(nscales);
dm.I1s.resize(nscales);
dm.u1s.resize(nscales);
dm.u2s.resize(nscales);
dm.u3s.resize(nscales);
I0.convertTo(dm.I0s[0], dm.I0s[0].depth(), I0.depth() == CV_8U ? 1.0 : 255.0);
I1.convertTo(dm.I1s[0], dm.I1s[0].depth(), I1.depth() == CV_8U ? 1.0 : 255.0);
dm.u1s[0].create(I0.size());
dm.u2s[0].create(I0.size());
if (use_gamma) dm.u3s[0].create(I0.size());
if (useInitialFlow)
{
Mat_<float> mv[] = { dm.u1s[0], dm.u2s[0] };
split(_flow.getMat(), mv);
}
dm.I1x_buf.create(I0.size());
dm.I1y_buf.create(I0.size());
dm.flowMap1_buf.create(I0.size());
dm.flowMap2_buf.create(I0.size());
dm.I1w_buf.create(I0.size());
dm.I1wx_buf.create(I0.size());
dm.I1wy_buf.create(I0.size());
dm.grad_buf.create(I0.size());
dm.rho_c_buf.create(I0.size());
dm.v1_buf.create(I0.size());
dm.v2_buf.create(I0.size());
dm.v3_buf.create(I0.size());
dm.p11_buf.create(I0.size());
dm.p12_buf.create(I0.size());
dm.p21_buf.create(I0.size());
dm.p22_buf.create(I0.size());
dm.p31_buf.create(I0.size());
dm.p32_buf.create(I0.size());
dm.div_p1_buf.create(I0.size());
dm.div_p2_buf.create(I0.size());
dm.div_p3_buf.create(I0.size());
dm.u1x_buf.create(I0.size());
dm.u1y_buf.create(I0.size());
dm.u2x_buf.create(I0.size());
dm.u2y_buf.create(I0.size());
dm.u3x_buf.create(I0.size());
dm.u3y_buf.create(I0.size());
// create the scales
for (int s = 1; s < nscales; ++s)
{
resize(dm.I0s[s - 1], dm.I0s[s], Size(), scaleStep, scaleStep, INTER_LINEAR);
resize(dm.I1s[s - 1], dm.I1s[s], Size(), scaleStep, scaleStep, INTER_LINEAR);
if (dm.I0s[s].cols < 16 || dm.I0s[s].rows < 16)
{
nscales = s;
break;
}
if (useInitialFlow)
{
resize(dm.u1s[s - 1], dm.u1s[s], Size(), scaleStep, scaleStep, INTER_LINEAR);
resize(dm.u2s[s - 1], dm.u2s[s], Size(), scaleStep, scaleStep, INTER_LINEAR);
multiply(dm.u1s[s], Scalar::all(scaleStep), dm.u1s[s]);
multiply(dm.u2s[s], Scalar::all(scaleStep), dm.u2s[s]);
}
else
{
dm.u1s[s].create(dm.I0s[s].size());
dm.u2s[s].create(dm.I0s[s].size());
}
if (use_gamma) dm.u3s[s].create(dm.I0s[s].size());
}
if (!useInitialFlow)
{
dm.u1s[nscales - 1].setTo(Scalar::all(0));
dm.u2s[nscales - 1].setTo(Scalar::all(0));
}
if (use_gamma) dm.u3s[nscales - 1].setTo(Scalar::all(0));
// pyramidal structure for computing the optical flow
for (int s = nscales - 1; s >= 0; --s)
{
// compute the optical flow at the current scale
procOneScale(dm.I0s[s], dm.I1s[s], dm.u1s[s], dm.u2s[s], dm.u3s[s]);
// if this was the last scale, finish now
if (s == 0)
break;
// otherwise, upsample the optical flow
// zoom the optical flow for the next finer scale
resize(dm.u1s[s], dm.u1s[s - 1], dm.I0s[s - 1].size(), 0, 0, INTER_LINEAR);
resize(dm.u2s[s], dm.u2s[s - 1], dm.I0s[s - 1].size(), 0, 0, INTER_LINEAR);
if (use_gamma) resize(dm.u3s[s], dm.u3s[s - 1], dm.I0s[s - 1].size(), 0, 0, INTER_LINEAR);
// scale the optical flow with the appropriate zoom factor (don't scale u3!)
multiply(dm.u1s[s - 1], Scalar::all(1 / scaleStep), dm.u1s[s - 1]);
multiply(dm.u2s[s - 1], Scalar::all(1 / scaleStep), dm.u2s[s - 1]);
}
Mat uxy[] = { dm.u1s[0], dm.u2s[0] };
merge(uxy, 2, _flow);
}
#ifdef HAVE_OPENCL
bool OpticalFlowDual_TVL1::calc_ocl(InputArray _I0, InputArray _I1, InputOutputArray _flow)
{
UMat I0 = _I0.getUMat();
UMat I1 = _I1.getUMat();
CV_Assert(I0.type() == CV_8UC1 || I0.type() == CV_32FC1);
CV_Assert(I0.size() == I1.size());
CV_Assert(I0.type() == I1.type());
CV_Assert(!useInitialFlow || (_flow.size() == I0.size() && _flow.type() == CV_32FC2));
CV_Assert(nscales > 0);
// allocate memory for the pyramid structure
dum.I0s.resize(nscales);
dum.I1s.resize(nscales);
dum.u1s.resize(nscales);
dum.u2s.resize(nscales);
//I0s_step == I1s_step
double alpha = I0.depth() == CV_8U ? 1.0 : 255.0;
I0.convertTo(dum.I0s[0], CV_32F, alpha);
I1.convertTo(dum.I1s[0], CV_32F, I1.depth() == CV_8U ? 1.0 : 255.0);
dum.u1s[0].create(I0.size(), CV_32FC1);
dum.u2s[0].create(I0.size(), CV_32FC1);
if (useInitialFlow)
{
std::vector<UMat> umv;
umv.push_back(dum.u1s[0]);
umv.push_back(dum.u2s[0]);
cv::split(_flow,umv);
}
dum.I1x_buf.create(I0.size(), CV_32FC1);
dum.I1y_buf.create(I0.size(), CV_32FC1);
dum.I1w_buf.create(I0.size(), CV_32FC1);
dum.I1wx_buf.create(I0.size(), CV_32FC1);
dum.I1wy_buf.create(I0.size(), CV_32FC1);
dum.grad_buf.create(I0.size(), CV_32FC1);
dum.rho_c_buf.create(I0.size(), CV_32FC1);
dum.p11_buf.create(I0.size(), CV_32FC1);
dum.p12_buf.create(I0.size(), CV_32FC1);
dum.p21_buf.create(I0.size(), CV_32FC1);
dum.p22_buf.create(I0.size(), CV_32FC1);
dum.diff_buf.create(I0.size(), CV_32FC1);
// create the scales
for (int s = 1; s < nscales; ++s)
{
resize(dum.I0s[s - 1], dum.I0s[s], Size(), scaleStep, scaleStep, INTER_LINEAR);
resize(dum.I1s[s - 1], dum.I1s[s], Size(), scaleStep, scaleStep, INTER_LINEAR);
if (dum.I0s[s].cols < 16 || dum.I0s[s].rows < 16)
{
nscales = s;
break;
}
if (useInitialFlow)
{
resize(dum.u1s[s - 1], dum.u1s[s], Size(), scaleStep, scaleStep, INTER_LINEAR);
resize(dum.u2s[s - 1], dum.u2s[s], Size(), scaleStep, scaleStep, INTER_LINEAR);
//scale by scale factor
multiply(dum.u1s[s], Scalar::all(scaleStep), dum.u1s[s]);
multiply(dum.u2s[s], Scalar::all(scaleStep), dum.u2s[s]);
}
}
// pyramidal structure for computing the optical flow
for (int s = nscales - 1; s >= 0; --s)
{
// compute the optical flow at the current scale
if (!OpticalFlowDual_TVL1::procOneScale_ocl(dum.I0s[s], dum.I1s[s], dum.u1s[s], dum.u2s[s]))
return false;
// if this was the last scale, finish now
if (s == 0)
break;
// zoom the optical flow for the next finer scale
resize(dum.u1s[s], dum.u1s[s - 1], dum.I0s[s - 1].size(), 0, 0, INTER_LINEAR);
resize(dum.u2s[s], dum.u2s[s - 1], dum.I0s[s - 1].size(), 0, 0, INTER_LINEAR);
// scale the optical flow with the appropriate zoom factor
multiply(dum.u1s[s - 1], Scalar::all(1 / scaleStep), dum.u1s[s - 1]);
multiply(dum.u2s[s - 1], Scalar::all(1 / scaleStep), dum.u2s[s - 1]);
}
std::vector<UMat> uxy;
uxy.push_back(dum.u1s[0]);
uxy.push_back(dum.u2s[0]);
merge(uxy, _flow);
return true;
}
#endif
////////////////////////////////////////////////////////////
// buildFlowMap
struct BuildFlowMapBody : ParallelLoopBody
{
void operator() (const Range& range) const;
Mat_<float> u1;
Mat_<float> u2;
mutable Mat_<float> map1;
mutable Mat_<float> map2;
};
void BuildFlowMapBody::operator() (const Range& range) const
{
for (int y = range.start; y < range.end; ++y)
{
const float* u1Row = u1[y];
const float* u2Row = u2[y];
float* map1Row = map1[y];
float* map2Row = map2[y];
for (int x = 0; x < u1.cols; ++x)
{
map1Row[x] = x + u1Row[x];
map2Row[x] = y + u2Row[x];
}
}
}
void buildFlowMap(const Mat_<float>& u1, const Mat_<float>& u2, Mat_<float>& map1, Mat_<float>& map2)
{
CV_DbgAssert( u2.size() == u1.size() );
CV_DbgAssert( map1.size() == u1.size() );
CV_DbgAssert( map2.size() == u1.size() );
BuildFlowMapBody body;
body.u1 = u1;
body.u2 = u2;
body.map1 = map1;
body.map2 = map2;
parallel_for_(Range(0, u1.rows), body);
}
////////////////////////////////////////////////////////////
// centeredGradient
struct CenteredGradientBody : ParallelLoopBody
{
void operator() (const Range& range) const;
Mat_<float> src;
mutable Mat_<float> dx;
mutable Mat_<float> dy;
};
void CenteredGradientBody::operator() (const Range& range) const
{
const int last_col = src.cols - 1;
for (int y = range.start; y < range.end; ++y)
{
const float* srcPrevRow = src[y - 1];
const float* srcCurRow = src[y];
const float* srcNextRow = src[y + 1];
float* dxRow = dx[y];
float* dyRow = dy[y];
for (int x = 1; x < last_col; ++x)
{
dxRow[x] = 0.5f * (srcCurRow[x + 1] - srcCurRow[x - 1]);
dyRow[x] = 0.5f * (srcNextRow[x] - srcPrevRow[x]);
}
}
}
void centeredGradient(const Mat_<float>& src, Mat_<float>& dx, Mat_<float>& dy)
{
CV_DbgAssert( src.rows > 2 && src.cols > 2 );
CV_DbgAssert( dx.size() == src.size() );
CV_DbgAssert( dy.size() == src.size() );
const int last_row = src.rows - 1;
const int last_col = src.cols - 1;
// compute the gradient on the center body of the image
{
CenteredGradientBody body;
body.src = src;
body.dx = dx;
body.dy = dy;
parallel_for_(Range(1, last_row), body);
}
// compute the gradient on the first and last rows
for (int x = 1; x < last_col; ++x)
{
dx(0, x) = 0.5f * (src(0, x + 1) - src(0, x - 1));
dy(0, x) = 0.5f * (src(1, x) - src(0, x));
dx(last_row, x) = 0.5f * (src(last_row, x + 1) - src(last_row, x - 1));
dy(last_row, x) = 0.5f * (src(last_row, x) - src(last_row - 1, x));
}
// compute the gradient on the first and last columns
for (int y = 1; y < last_row; ++y)
{
dx(y, 0) = 0.5f * (src(y, 1) - src(y, 0));
dy(y, 0) = 0.5f * (src(y + 1, 0) - src(y - 1, 0));
dx(y, last_col) = 0.5f * (src(y, last_col) - src(y, last_col - 1));
dy(y, last_col) = 0.5f * (src(y + 1, last_col) - src(y - 1, last_col));
}
// compute the gradient at the four corners
dx(0, 0) = 0.5f * (src(0, 1) - src(0, 0));
dy(0, 0) = 0.5f * (src(1, 0) - src(0, 0));
dx(0, last_col) = 0.5f * (src(0, last_col) - src(0, last_col - 1));
dy(0, last_col) = 0.5f * (src(1, last_col) - src(0, last_col));
dx(last_row, 0) = 0.5f * (src(last_row, 1) - src(last_row, 0));
dy(last_row, 0) = 0.5f * (src(last_row, 0) - src(last_row - 1, 0));
dx(last_row, last_col) = 0.5f * (src(last_row, last_col) - src(last_row, last_col - 1));
dy(last_row, last_col) = 0.5f * (src(last_row, last_col) - src(last_row - 1, last_col));
}
////////////////////////////////////////////////////////////
// forwardGradient
struct ForwardGradientBody : ParallelLoopBody
{
void operator() (const Range& range) const;
Mat_<float> src;
mutable Mat_<float> dx;
mutable Mat_<float> dy;
};
void ForwardGradientBody::operator() (const Range& range) const
{
const int last_col = src.cols - 1;
for (int y = range.start; y < range.end; ++y)
{
const float* srcCurRow = src[y];
const float* srcNextRow = src[y + 1];
float* dxRow = dx[y];
float* dyRow = dy[y];
for (int x = 0; x < last_col; ++x)
{
dxRow[x] = srcCurRow[x + 1] - srcCurRow[x];
dyRow[x] = srcNextRow[x] - srcCurRow[x];
}
}
}
void forwardGradient(const Mat_<float>& src, Mat_<float>& dx, Mat_<float>& dy)
{
CV_DbgAssert( src.rows > 2 && src.cols > 2 );
CV_DbgAssert( dx.size() == src.size() );
CV_DbgAssert( dy.size() == src.size() );
const int last_row = src.rows - 1;
const int last_col = src.cols - 1;
// compute the gradient on the central body of the image
{
ForwardGradientBody body;
body.src = src;
body.dx = dx;
body.dy = dy;
parallel_for_(Range(0, last_row), body);
}
// compute the gradient on the last row
for (int x = 0; x < last_col; ++x)
{
dx(last_row, x) = src(last_row, x + 1) - src(last_row, x);
dy(last_row, x) = 0.0f;
}
// compute the gradient on the last column
for (int y = 0; y < last_row; ++y)
{
dx(y, last_col) = 0.0f;
dy(y, last_col) = src(y + 1, last_col) - src(y, last_col);
}
dx(last_row, last_col) = 0.0f;
dy(last_row, last_col) = 0.0f;
}
////////////////////////////////////////////////////////////
// divergence
struct DivergenceBody : ParallelLoopBody
{
void operator() (const Range& range) const;
Mat_<float> v1;
Mat_<float> v2;
mutable Mat_<float> div;
};
void DivergenceBody::operator() (const Range& range) const
{
for (int y = range.start; y < range.end; ++y)
{
const float* v1Row = v1[y];
const float* v2PrevRow = v2[y - 1];
const float* v2CurRow = v2[y];
float* divRow = div[y];
for(int x = 1; x < v1.cols; ++x)
{
const float v1x = v1Row[x] - v1Row[x - 1];
const float v2y = v2CurRow[x] - v2PrevRow[x];
divRow[x] = v1x + v2y;
}
}
}
void divergence(const Mat_<float>& v1, const Mat_<float>& v2, Mat_<float>& div)
{
CV_DbgAssert( v1.rows > 2 && v1.cols > 2 );
CV_DbgAssert( v2.size() == v1.size() );
CV_DbgAssert( div.size() == v1.size() );
{
DivergenceBody body;
body.v1 = v1;
body.v2 = v2;
body.div = div;
parallel_for_(Range(1, v1.rows), body);
}
// compute the divergence on the first row
for(int x = 1; x < v1.cols; ++x)
div(0, x) = v1(0, x) - v1(0, x - 1) + v2(0, x);
// compute the divergence on the first column
for (int y = 1; y < v1.rows; ++y)
div(y, 0) = v1(y, 0) + v2(y, 0) - v2(y - 1, 0);
div(0, 0) = v1(0, 0) + v2(0, 0);
}
////////////////////////////////////////////////////////////
// calcGradRho
struct CalcGradRhoBody : ParallelLoopBody
{
void operator() (const Range& range) const;
Mat_<float> I0;
Mat_<float> I1w;
Mat_<float> I1wx;
Mat_<float> I1wy;
Mat_<float> u1;
Mat_<float> u2;
mutable Mat_<float> grad;
mutable Mat_<float> rho_c;
};
void CalcGradRhoBody::operator() (const Range& range) const
{
for (int y = range.start; y < range.end; ++y)
{
const float* I0Row = I0[y];
const float* I1wRow = I1w[y];
const float* I1wxRow = I1wx[y];
const float* I1wyRow = I1wy[y];
const float* u1Row = u1[y];
const float* u2Row = u2[y];
float* gradRow = grad[y];
float* rhoRow = rho_c[y];
for (int x = 0; x < I0.cols; ++x)
{
const float Ix2 = I1wxRow[x] * I1wxRow[x];
const float Iy2 = I1wyRow[x] * I1wyRow[x];
// store the |Grad(I1)|^2
gradRow[x] = Ix2 + Iy2;
// compute the constant part of the rho function
rhoRow[x] = (I1wRow[x] - I1wxRow[x] * u1Row[x] - I1wyRow[x] * u2Row[x] - I0Row[x]);
}
}
}
void calcGradRho(const Mat_<float>& I0, const Mat_<float>& I1w, const Mat_<float>& I1wx, const Mat_<float>& I1wy, const Mat_<float>& u1, const Mat_<float>& u2,
Mat_<float>& grad, Mat_<float>& rho_c)
{
CV_DbgAssert( I1w.size() == I0.size() );
CV_DbgAssert( I1wx.size() == I0.size() );
CV_DbgAssert( I1wy.size() == I0.size() );
CV_DbgAssert( u1.size() == I0.size() );
CV_DbgAssert( u2.size() == I0.size() );
CV_DbgAssert( grad.size() == I0.size() );
CV_DbgAssert( rho_c.size() == I0.size() );
CalcGradRhoBody body;
body.I0 = I0;
body.I1w = I1w;
body.I1wx = I1wx;
body.I1wy = I1wy;
body.u1 = u1;
body.u2 = u2;
body.grad = grad;
body.rho_c = rho_c;
parallel_for_(Range(0, I0.rows), body);
}
////////////////////////////////////////////////////////////
// estimateV
struct EstimateVBody : ParallelLoopBody
{
void operator() (const Range& range) const;
Mat_<float> I1wx;
Mat_<float> I1wy;
Mat_<float> u1;
Mat_<float> u2;
Mat_<float> u3;
Mat_<float> grad;
Mat_<float> rho_c;
mutable Mat_<float> v1;
mutable Mat_<float> v2;
mutable Mat_<float> v3;
float l_t;
float gamma;
};
void EstimateVBody::operator() (const Range& range) const
{
bool use_gamma = gamma != 0;
for (int y = range.start; y < range.end; ++y)
{
const float* I1wxRow = I1wx[y];
const float* I1wyRow = I1wy[y];
const float* u1Row = u1[y];
const float* u2Row = u2[y];
const float* u3Row = use_gamma?u3[y]:NULL;
const float* gradRow = grad[y];
const float* rhoRow = rho_c[y];
float* v1Row = v1[y];
float* v2Row = v2[y];
float* v3Row = use_gamma ? v3[y]:NULL;
for (int x = 0; x < I1wx.cols; ++x)
{
const float rho = use_gamma ? rhoRow[x] + (I1wxRow[x] * u1Row[x] + I1wyRow[x] * u2Row[x]) + gamma * u3Row[x] :
rhoRow[x] + (I1wxRow[x] * u1Row[x] + I1wyRow[x] * u2Row[x]);