-
Notifications
You must be signed in to change notification settings - Fork 6
Expand file tree
/
Copy pathslic.cpp
More file actions
1042 lines (925 loc) · 45.8 KB
/
slic.cpp
File metadata and controls
1042 lines (925 loc) · 45.8 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
#include "stdafx.h"
#include <CL/cl.h>
#include <opencv2/opencv.hpp>
// #include <opencv2/ximgproc.hpp>
#include <iostream>
#include <vector>
#include <fstream>
#include <sstream>
#include <chrono>
#include <stdexcept>
#include <random>
#include "slic.h"
// OpenCL kernel source code
const char* kernelSource1 = R"(
// Convert RGB to LAB color space
float3 rgb2lab(float r, float g, float b) {
float3 lab;
if (r > 0.04045f) r = pow((r + 0.055f) / 1.055f, 2.4f);
else r = r / 12.92f;
if (g > 0.04045f) g = pow((g + 0.055f) / 1.055f, 2.4f);
else g = g / 12.92f;
if (b > 0.04045f) b = pow((b + 0.055f) / 1.055f, 2.4f);
else b = b / 12.92f;
r *= 100.0f;
g *= 100.0f;
b *= 100.0f;
float x = r * 0.4124f + g * 0.3576f + b * 0.1805f;
float y = r * 0.2126f + g * 0.7152f + b * 0.0722f;
float z = r * 0.0193f + g * 0.1192f + b * 0.9505f;
x = x / 95.047f;
y = y / 100.0f;
z = z / 108.883f;
if (x > 0.008856f) x = pow(x, 1.0f / 3.0f);
else x = (7.787f * x) + (16.0f / 116.0f);
if (y > 0.008856f) y = pow(y, 1.0f / 3.0f);
else y = (7.787f * y) + (16.0f / 116.0f);
if (z > 0.008856f) z = pow(z, 1.0f / 3.0f);
else z = (7.787f * z) + (16.0f / 116.0f);
lab.x = (116.0f * y) - 16.0f; // L
lab.y = 500.0f * (x - y); // a
lab.z = 200.0f * (y - z); // b
return lab;
}
)";
const char* kernelSource2 = R"(
// Initialize cluster centers
__kernel void initClusters(
__global const float* input,
__global float* clusters,
int width, int height,
int numSuperpixels) {
int clusterIdx = get_global_id(0);
if (clusterIdx >= numSuperpixels) return;
// Calculate grid step
float stepX = (float)width / sqrt((float)numSuperpixels);
float stepY = (float)height / sqrt((float)numSuperpixels);
// Calculate grid position
int gridY = (int)floor(clusterIdx / floor((float)width / stepX));
int gridX = clusterIdx - (gridY * floor((float)width / stepX));
// Calculate center position
int centerX = (int)(stepX * (gridX + 0.5f));
int centerY = (int)(stepY * (gridY + 0.5f));
// Clamp to image boundaries
centerX = min(max(centerX, 0), width - 1);
centerY = min(max(centerY, 0), height - 1);
// Get center pixel color
int pixIdx = (centerY * width + centerX) * 3;
float r = input[pixIdx];
float g = input[pixIdx + 1];
float b = input[pixIdx + 2];
// Convert to LAB
float3 lab = rgb2lab(r, g, b);
// Store cluster center [x, y, l, a, b]
clusters[clusterIdx * 5 + 0] = (float)centerX;
clusters[clusterIdx * 5 + 1] = (float)centerY;
clusters[clusterIdx * 5 + 2] = lab.x;
clusters[clusterIdx * 5 + 3] = lab.y;
clusters[clusterIdx * 5 + 4] = lab.z;
}
)";
const char* kernelSource3 = R"(
// Assign pixels to clusters
__kernel void assignPixels(
__global const float* input,
__global const float* clusters,
__global int* labels,
int width, int height,
int numSuperpixels,
int regionSize,
float compactness) {
int x = get_global_id(0);
int y = get_global_id(1);
if (x >= width || y >= height) return;
int pixIdx = (y * width + x) * 3;
float3 lab = rgb2lab(input[pixIdx], input[pixIdx + 1], input[pixIdx + 2]);
// Initialize distance to a large value
float minDist = 1.0e10f;
int nearestCluster = -1;
// Calculate grid step
float stepX = (float)width / sqrt((float)numSuperpixels);
float stepY = (float)height / sqrt((float)numSuperpixels);
int searchRegionSize = (int)(max(stepX, stepY) * 2.0f);
// Calculate grid position of the pixel
int gridX = (int)floor(x / stepX);
int gridY = (int)floor(y / stepY);
// Search in a 2x2 grid neighborhood
for (int i = -1; i <= 1; i++) {
for (int j = -1; j <= 1; j++) {
int gridIdxX = gridX + i;
int gridIdxY = gridY + j;
// Skip if outside grid boundaries
if (gridIdxX < 0 || gridIdxY < 0 ||
gridIdxX >= (int)ceil((float)width / stepX) ||
gridIdxY >= (int)ceil((float)height / stepY))
continue;
// Calculate cluster index
int clusterId = gridIdxY * (int)floor((float)width / stepX) + gridIdxX;
if (clusterId >= numSuperpixels) continue;
// Get cluster center
float cX = clusters[clusterId * 5 + 0];
float cY = clusters[clusterId * 5 + 1];
// Skip if outside search region
if (fabs(cX - x) > searchRegionSize || fabs(cY - y) > searchRegionSize)
continue;
float cL = clusters[clusterId * 5 + 2];
float cA = clusters[clusterId * 5 + 3];
float cB = clusters[clusterId * 5 + 4];
// Calculate color distance (CIELAB)
float distL = cL - lab.x;
float distA = cA - lab.y;
float distB = cB - lab.z;
float distColor = sqrt(distL*distL + distA*distA + distB*distB);
// Calculate spatial distance
float distX = cX - x;
float distY = cY - y;
float distSpace = sqrt(distX*distX + distY*distY);
// SLIC distance metric: D = distColor + (compactness/regionSize) * distSpace
float dist = distColor + (compactness / (float)regionSize) * distSpace;
// Update nearest cluster
if (dist < minDist) {
minDist = dist;
nearestCluster = clusterId;
}
}
}
// Assign pixel to nearest cluster
if (nearestCluster >= 0) {
labels[y * width + x] = nearestCluster;
}
}
)";
const char* kernelSource4 = R"(
// Update cluster centers
__kernel void updateClusters(
__global const float* input,
__global float* clusters,
__global const int* labels,
int width, int height,
int numSuperpixels) {
int idx = get_global_id(0);
int x = idx % width;
int y = idx / width;
if (x >= width || y >= height) return;
int pixIdx = (y * width + x) * 3;
int label = labels[y * width + x];
if (label >= 0 && label < numSuperpixels) {
// Convert RGB to LAB
float3 lab = rgb2lab(input[pixIdx], input[pixIdx + 1], input[pixIdx + 2]);
// Use local memory for accumulation
__local float localAccum[5];
if (get_local_id(0) == 0) {
for (int i = 0; i < 5; i++) {
localAccum[i] = 0.0f;
}
}
barrier(CLK_LOCAL_MEM_FENCE);
// Accumulate values
localAccum[0] += (float)x;
localAccum[1] += (float)y;
localAccum[2] += lab.x;
localAccum[3] += lab.y;
localAccum[4] += lab.z;
barrier(CLK_LOCAL_MEM_FENCE);
// Write back to global memory
if (get_local_id(0) == 0) {
for (int i = 0; i < 5; i++) {
clusters[label * 5 + i] = localAccum[i];
}
}
}
}
)";
const char* kernelSource5 = R"(
#define SCALE 1000
__kernel void accumulateClusters(
__global const float* input,
__global const int* labels,
__global int* accum, // [numClusters * 5]
__global int* counts, // [numClusters]
int width, int height)
{
int idx = get_global_id(0);
int x = idx % width;
int y = idx / width;
int label = labels[idx];
float3 lab = rgb2lab(input[idx*3], input[idx*3+1], input[idx*3+2]);
atomic_add(&accum[label*5+0], (int)(x * SCALE));
atomic_add(&accum[label*5+1], (int)(y * SCALE));
atomic_add(&accum[label*5+2], (int)(lab.x * SCALE));
atomic_add(&accum[label*5+3], (int)(lab.y * SCALE));
atomic_add(&accum[label*5+4], (int)(lab.z * SCALE));
atomic_inc(&counts[label]);
}
)";
const char* kernelSource6 = R"(
#define SCALE 1000
__kernel void finalizeClusters(
__global int* accum,
__global int* counts,
__global float* clusters,
int numClusters)
{
int i = get_global_id(0);
if (counts[i] > 0) {
clusters[i*5+0] = (float)accum[i*5+0] / (counts[i] * SCALE);
clusters[i*5+1] = (float)accum[i*5+1] / (counts[i] * SCALE);
clusters[i*5+2] = (float)accum[i*5+2] / (counts[i] * SCALE);
clusters[i*5+3] = (float)accum[i*5+3] / (counts[i] * SCALE);
clusters[i*5+4] = (float)accum[i*5+4] / (counts[i] * SCALE);
}
}
)";
const char* kernelSource7 = R"(
float atomic_add_float(__global float* addr, float val) {
__global uint* address_as_uint = (__global uint*)addr;
uint old = *address_as_uint;
uint assumed;
do {
assumed = old;
old = atomic_cmpxchg(
address_as_uint,
assumed,
as_uint(val + as_float(assumed)));
} while (assumed != old);
return as_float(old);
}
float atomic_add_float_local(__local float* addr, float val) {
__local uint* address_as_uint = (__local uint*)addr;
uint old = *address_as_uint;
uint assumed;
do {
assumed = old;
old = atomic_cmpxchg(
address_as_uint,
assumed,
as_uint(val + as_float(assumed)));
} while (assumed != old);
return as_float(old);
}
__kernel void superpixel_atomic_average(
__global const int* labels,
__global const float* image, // [width*height]
__global float* sum, // [numSuperpixels]
__global int* count, // [numSuperpixels]
int numPixels)
{
int idx = get_global_id(0);
if (idx >= numPixels) return;
int label = labels[idx];
float value = image[idx];
atomic_add_float(&sum[label], value);
atomic_inc(&count[label]);
}
__kernel void superpixel_efficient_average(
__global const int* labels,
__global const float* image,
__global float* sum, // [numSuperpixels]
__global int* count, // [numSuperpixels]
__local float* local_sums, // Local memory for work group
__local int* local_counts, // Local memory for work group
int numSuperpixels,
int numPixels)
{
// Get indices
int gid = get_global_id(0);
int lid = get_local_id(0);
int local_size = get_local_size(0);
// Initialize local memory
for (int i = lid; i < numSuperpixels; i += local_size) {
local_sums[i] = 0.0f;
local_counts[i] = 0;
}
barrier(CLK_LOCAL_MEM_FENCE);
// First phase: Accumulate in local memory
if (gid < numPixels) {
int label = labels[gid];
float value = image[gid];
// Using atomic operations on local memory (much faster than global)
atomic_add_float_local(&local_sums[label], value);
atomic_inc(&local_counts[label]);
}
barrier(CLK_LOCAL_MEM_FENCE);
// Second phase: Accumulate from local to global
// Only one thread per label in the work group will do this
for (int label = lid; label < numSuperpixels; label += local_size) {
if (local_counts[label] > 0) {
atomic_add_float(&sum[label], local_sums[label]);
atomic_add(&count[label], local_counts[label]);
}
}
}
__kernel void compute_averages(
__global float* sum, // [numSuperpixels]
__global const int* count, // [numSuperpixels]
__global float* averages, // [numSuperpixels]
int numSuperpixels)
{
int idx = get_global_id(0);
if (idx >= numSuperpixels) return;
if (count[idx] > 0) {
averages[idx] = sum[idx] / count[idx];
}
}
)";
const char* kernelSource8 = R"(
__kernel void assign_superpixel_average(
__global const int* labels, // [width*height]
__global const float* averages, // [numSuperpixels]
__global float* out_image, // [width*height]
int numPixels)
{
int idx = get_global_id(0);
if (idx >= numPixels) return;
int label = labels[idx];
out_image[idx] = averages[label];
}
)";
const char* kernelSource9 = R"(
__kernel void superpixel_efficient_average2(
__global const int* labels,
__global const float* image1,
__global const float* image2,
__global float* sum1,
__global float* sum2,
__global int* count1,
__global int* count2,
__local float* local_sums1,
__local float* local_sums2,
__local int* local_counts1,
__local int* local_counts2,
int numSuperpixels,
int numPixels)
{
// Get indices
int gid = get_global_id(0);
int lid = get_local_id(0);
int local_size = get_local_size(0);
// Initialize local memory
for (int i = lid; i < numSuperpixels; i += local_size) {
local_sums1[i] = 0.0f;
local_sums2[i] = 0.0f;
local_counts1[i] = 0;
local_counts2[i] = 0;
}
barrier(CLK_LOCAL_MEM_FENCE);
// First phase: Accumulate in local memory
if (gid < numPixels) {
int label = labels[gid];
float value1 = image1[gid];
float value2 = image2[gid];
// Using atomic operations on local memory
atomic_add_float_local(&local_sums1[label], value1);
atomic_add_float_local(&local_sums2[label], value2);
atomic_inc(&local_counts1[label]);
atomic_inc(&local_counts2[label]);
}
barrier(CLK_LOCAL_MEM_FENCE);
// Second phase: Accumulate from local to global
for (int label = lid; label < numSuperpixels; label += local_size) {
if (local_counts1[label] > 0) {
atomic_add_float(&sum1[label], local_sums1[label]);
atomic_add_float(&sum2[label], local_sums2[label]);
atomic_add(&count1[label], local_counts1[label]);
atomic_add(&count2[label], local_counts2[label]);
}
}
}
__kernel void compute_averages2(
__global float* sum1,
__global float* sum2,
__global const int* count1,
__global const int* count2,
__global float* averages1,
__global float* averages2,
int numSuperpixels)
{
int idx = get_global_id(0);
if (idx >= numSuperpixels) return;
if (count1[idx] > 0) {
averages1[idx] = sum1[idx] / count1[idx];
averages2[idx] = sum2[idx] / count2[idx];
}
}
__kernel void assign_superpixel_average2(
__global const int* labels,
__global const float* averages1,
__global const float* averages2,
__global float* out_image1,
__global float* out_image2,
int numPixels)
{
int idx = get_global_id(0);
if (idx >= numPixels) return;
int label = labels[idx];
out_image1[idx] = averages1[label];
out_image2[idx] = averages2[label];
}
)";
// Helper function to load OpenCL kernel from string
std::string SLICSuperpixels::loadKernelSource() {
std::string source;
source += kernelSource1;
source += kernelSource2;
source += kernelSource3;
source += kernelSource4;
source += kernelSource5;
source += kernelSource6;
source += kernelSource7;
source += kernelSource8;
source += kernelSource9;
return source;
}
// Initialize OpenCL
void SLICSuperpixels::initOpenCL() {
cl_int err;
std::string kernelSource = loadKernelSource();
const char* source = kernelSource.c_str();
size_t sourceSize = kernelSource.length();
program = clCreateProgramWithSource(context, 1, &source, &sourceSize, &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create program");
err = clBuildProgram(program, 1, &device, nullptr, nullptr, nullptr);
if (err != CL_SUCCESS) {
size_t logSize;
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, 0, nullptr, &logSize);
std::vector<char> log(logSize);
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, logSize, log.data(), nullptr);
std::cerr << "Build log: " << log.data() << std::endl;
throw std::runtime_error("Failed to build program");
}
initKernel = clCreateKernel(program, "initClusters", &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create init kernel");
assignKernel = clCreateKernel(program, "assignPixels", &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create assign kernel");
updateKernel = clCreateKernel(program, "updateClusters", &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create update kernel");
accumulateKernel = clCreateKernel(program, "accumulateClusters", &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create accumulate kernel");
finalizeKernel = clCreateKernel(program, "finalizeClusters", &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create finalize kernel");
// Create kernels for single map case
averageKernel = clCreateKernel(program, "superpixel_efficient_average", &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create average kernel");
computeKernel = clCreateKernel(program, "compute_averages", &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create compute kernel");
assignAvgKernel = clCreateKernel(program, "assign_superpixel_average", &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create assign average kernel");
// Create kernels for two map case
averageKernel2 = clCreateKernel(program, "superpixel_efficient_average2", &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create average2 kernel");
computeKernel2 = clCreateKernel(program, "compute_averages2", &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create compute2 kernel");
assignAvgKernel2 = clCreateKernel(program, "assign_superpixel_average2", &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create assign average2 kernel");
}
// Clean up OpenCL resources
void SLICSuperpixels::CleanupOpenCL() {
if (updateKernel) clReleaseKernel(updateKernel);
if (assignKernel) clReleaseKernel(assignKernel);
if (initKernel) clReleaseKernel(initKernel);
if (finalizeKernel) clReleaseKernel(finalizeKernel);
if (accumulateKernel) clReleaseKernel(accumulateKernel);
// Release single map kernels
if (averageKernel) clReleaseKernel(averageKernel);
if (computeKernel) clReleaseKernel(computeKernel);
if (assignAvgKernel) clReleaseKernel(assignAvgKernel);
// Release two-map kernels
if (averageKernel2) clReleaseKernel(averageKernel2);
if (computeKernel2) clReleaseKernel(computeKernel2);
if (assignAvgKernel2) clReleaseKernel(assignAvgKernel2);
if (program) clReleaseProgram(program);
}
SLICSuperpixels::SLICSuperpixels(cl_context context, cl_command_queue queue, cl_device_id device, float compactness, int numSuperpixels, int iterations)
: context(context), queue(queue), device(device), compactness(compactness), numSuperpixels(numSuperpixels), iterations(iterations) {
initOpenCL();
}
SLICSuperpixels::~SLICSuperpixels() {
releaseBuffers();
CleanupOpenCL();
}
// Allocate all needed buffers except d_image
void SLICSuperpixels::allocateBuffers(int width_, int height_, bool processTwoMaps_) {
width = width_;
height = height_;
regionSize = static_cast<int>(std::round(std::sqrt((width * height) / (float)numSuperpixels)));
globalSize = numSuperpixels;
numPixels = width * height;
globalSize2[0] = static_cast<size_t>(width);
globalSize2[1] = static_cast<size_t>(height);
globalSize3 = width * height;
processTwoMaps = processTwoMaps_;
cl_int err;
d_clusters = clCreateBuffer(context, CL_MEM_READ_WRITE,
numSuperpixels * 5 * sizeof(float), nullptr, &err);
d_accum = clCreateBuffer(context, CL_MEM_READ_WRITE, numSuperpixels * 5 * sizeof(int), nullptr, &err);
d_counts = clCreateBuffer(context, CL_MEM_READ_WRITE, numSuperpixels * sizeof(int), nullptr, &err);
if (processTwoMaps) {
// Buffers for two-map case
d_value_sum_per_superpixel1 = clCreateBuffer(context, CL_MEM_READ_WRITE, numSuperpixels * sizeof(float), nullptr, &err);
d_value_sum_per_superpixel2 = clCreateBuffer(context, CL_MEM_READ_WRITE, numSuperpixels * sizeof(float), nullptr, &err);
d_pixel_counts_per_superpixel1 = clCreateBuffer(context, CL_MEM_READ_WRITE, numSuperpixels * sizeof(int), nullptr, &err);
d_pixel_counts_per_superpixel2 = clCreateBuffer(context, CL_MEM_READ_WRITE, numSuperpixels * sizeof(int), nullptr, &err);
d_averages1 = clCreateBuffer(context, CL_MEM_READ_WRITE, numSuperpixels * sizeof(float), nullptr, &err);
d_averages2 = clCreateBuffer(context, CL_MEM_READ_WRITE, numSuperpixels * sizeof(float), nullptr, &err);
} else {
// Buffers for single map case
d_pixel_counts_per_superpixel = clCreateBuffer(context, CL_MEM_READ_WRITE, numPixels * sizeof(int), nullptr, &err);
d_value_sum_per_superpixel = clCreateBuffer(context, CL_MEM_READ_WRITE, numSuperpixels * sizeof(float), nullptr, &err);
d_averages = clCreateBuffer(context, CL_MEM_READ_WRITE, numSuperpixels * sizeof(float), nullptr, &err);
}
// Zero initialize all buffers
std::vector<int> zeroAccum(numSuperpixels * 5, 0);
std::vector<int> zeroCounts(numSuperpixels, 0);
std::vector<int> zeroPixelCounts(numPixels, 0);
std::vector<float> zeroValueSum(numSuperpixels, 0.0f);
std::vector<float> zeroAverages(numSuperpixels, 0.0f);
err = clEnqueueWriteBuffer(queue, d_accum, CL_TRUE, 0, numSuperpixels * 5 * sizeof(int), zeroAccum.data(), 0, nullptr, nullptr);
err |= clEnqueueWriteBuffer(queue, d_counts, CL_TRUE, 0, numSuperpixels * sizeof(int), zeroCounts.data(), 0, nullptr, nullptr);
if (processTwoMaps) {
// Initialize two-map buffers
err |= clEnqueueWriteBuffer(queue, d_value_sum_per_superpixel1, CL_TRUE, 0, numSuperpixels * sizeof(float), zeroValueSum.data(), 0, nullptr, nullptr);
err |= clEnqueueWriteBuffer(queue, d_value_sum_per_superpixel2, CL_TRUE, 0, numSuperpixels * sizeof(float), zeroValueSum.data(), 0, nullptr, nullptr);
err |= clEnqueueWriteBuffer(queue, d_pixel_counts_per_superpixel1, CL_TRUE, 0, numSuperpixels * sizeof(int), zeroCounts.data(), 0, nullptr, nullptr);
err |= clEnqueueWriteBuffer(queue, d_pixel_counts_per_superpixel2, CL_TRUE, 0, numSuperpixels * sizeof(int), zeroCounts.data(), 0, nullptr, nullptr);
err |= clEnqueueWriteBuffer(queue, d_averages1, CL_TRUE, 0, numSuperpixels * sizeof(float), zeroAverages.data(), 0, nullptr, nullptr);
err |= clEnqueueWriteBuffer(queue, d_averages2, CL_TRUE, 0, numSuperpixels * sizeof(float), zeroAverages.data(), 0, nullptr, nullptr);
} else {
// Initialize single map buffers
err |= clEnqueueWriteBuffer(queue, d_pixel_counts_per_superpixel, CL_TRUE, 0, numPixels * sizeof(int), zeroPixelCounts.data(), 0, nullptr, nullptr);
err |= clEnqueueWriteBuffer(queue, d_value_sum_per_superpixel, CL_TRUE, 0, numSuperpixels * sizeof(float), zeroValueSum.data(), 0, nullptr, nullptr);
err |= clEnqueueWriteBuffer(queue, d_averages, CL_TRUE, 0, numSuperpixels * sizeof(float), zeroAverages.data(), 0, nullptr, nullptr);
}
if (err != CL_SUCCESS) {
throw std::runtime_error("Failed to initialize buffers");
}
}
// Set all kernel arguments, using external d_image
void SLICSuperpixels::setKernelArgs(cl_mem d_image, cl_mem d_labels, cl_mem d_opticalFlowMap1, cl_mem d_opticalFlowMap2, cl_mem d_avgImage1, cl_mem d_avgImage2) {
cl_int err = 0;
// initKernel
err = clSetKernelArg(initKernel, 0, sizeof(cl_mem), &d_image);
err |= clSetKernelArg(initKernel, 1, sizeof(cl_mem), &d_clusters);
err |= clSetKernelArg(initKernel, 2, sizeof(int), &width);
err |= clSetKernelArg(initKernel, 3, sizeof(int), &height);
err |= clSetKernelArg(initKernel, 4, sizeof(int), &numSuperpixels);
// assignKernel
err = clSetKernelArg(assignKernel, 0, sizeof(cl_mem), &d_image);
err |= clSetKernelArg(assignKernel, 1, sizeof(cl_mem), &d_clusters);
err |= clSetKernelArg(assignKernel, 2, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(assignKernel, 3, sizeof(int), &width);
err |= clSetKernelArg(assignKernel, 4, sizeof(int), &height);
err |= clSetKernelArg(assignKernel, 5, sizeof(int), &numSuperpixels);
err |= clSetKernelArg(assignKernel, 6, sizeof(int), ®ionSize);
err |= clSetKernelArg(assignKernel, 7, sizeof(float), &compactness);
// accumulateKernel
err = clSetKernelArg(accumulateKernel, 0, sizeof(cl_mem), &d_image);
err |= clSetKernelArg(accumulateKernel, 1, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(accumulateKernel, 2, sizeof(cl_mem), &d_accum);
err |= clSetKernelArg(accumulateKernel, 3, sizeof(cl_mem), &d_counts);
err |= clSetKernelArg(accumulateKernel, 4, sizeof(int), &width);
err |= clSetKernelArg(accumulateKernel, 5, sizeof(int), &height);
// finalizeKernel
err = clSetKernelArg(finalizeKernel, 0, sizeof(cl_mem), &d_accum);
err |= clSetKernelArg(finalizeKernel, 1, sizeof(cl_mem), &d_counts);
err |= clSetKernelArg(finalizeKernel, 2, sizeof(cl_mem), &d_clusters);
err |= clSetKernelArg(finalizeKernel, 3, sizeof(int), &numSuperpixels);
if (processTwoMaps) {
// averageKernel2
size_t localMemSize1 = numSuperpixels * sizeof(float);
size_t localMemSize2 = numSuperpixels * sizeof(float);
size_t localCountSize1 = numSuperpixels * sizeof(int);
size_t localCountSize2 = numSuperpixels * sizeof(int);
err = clSetKernelArg(averageKernel2, 0, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(averageKernel2, 1, sizeof(cl_mem), &d_opticalFlowMap1);
err |= clSetKernelArg(averageKernel2, 2, sizeof(cl_mem), &d_opticalFlowMap2);
err |= clSetKernelArg(averageKernel2, 3, sizeof(cl_mem), &d_value_sum_per_superpixel1);
err |= clSetKernelArg(averageKernel2, 4, sizeof(cl_mem), &d_value_sum_per_superpixel2);
err |= clSetKernelArg(averageKernel2, 5, sizeof(cl_mem), &d_pixel_counts_per_superpixel1);
err |= clSetKernelArg(averageKernel2, 6, sizeof(cl_mem), &d_pixel_counts_per_superpixel2);
err |= clSetKernelArg(averageKernel2, 7, localMemSize1, nullptr); // Local memory for sums1
err |= clSetKernelArg(averageKernel2, 8, localMemSize2, nullptr); // Local memory for sums2
err |= clSetKernelArg(averageKernel2, 9, localCountSize1, nullptr); // Local memory for counts1
err |= clSetKernelArg(averageKernel2, 10, localCountSize2, nullptr); // Local memory for counts2
err |= clSetKernelArg(averageKernel2, 11, sizeof(int), &numSuperpixels);
err |= clSetKernelArg(averageKernel2, 12, sizeof(int), &numPixels);
// computeKernel2
err |= clSetKernelArg(computeKernel2, 0, sizeof(cl_mem), &d_value_sum_per_superpixel1);
err |= clSetKernelArg(computeKernel2, 1, sizeof(cl_mem), &d_value_sum_per_superpixel2);
err |= clSetKernelArg(computeKernel2, 2, sizeof(cl_mem), &d_pixel_counts_per_superpixel1);
err |= clSetKernelArg(computeKernel2, 3, sizeof(cl_mem), &d_pixel_counts_per_superpixel2);
err |= clSetKernelArg(computeKernel2, 4, sizeof(cl_mem), &d_averages1);
err |= clSetKernelArg(computeKernel2, 5, sizeof(cl_mem), &d_averages2);
err |= clSetKernelArg(computeKernel2, 6, sizeof(int), &numSuperpixels);
// assignAvgKernel2
err |= clSetKernelArg(assignAvgKernel2, 0, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(assignAvgKernel2, 1, sizeof(cl_mem), &d_averages1);
err |= clSetKernelArg(assignAvgKernel2, 2, sizeof(cl_mem), &d_averages2);
err |= clSetKernelArg(assignAvgKernel2, 3, sizeof(cl_mem), &d_avgImage1);
err |= clSetKernelArg(assignAvgKernel2, 4, sizeof(cl_mem), &d_avgImage2);
err |= clSetKernelArg(assignAvgKernel2, 5, sizeof(int), &numPixels);
} else {
// averageKernel
size_t localMemSize = numSuperpixels * sizeof(float);
size_t localCountSize = numSuperpixels * sizeof(int);
err = clSetKernelArg(averageKernel, 0, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(averageKernel, 1, sizeof(cl_mem), &d_opticalFlowMap1);
err |= clSetKernelArg(averageKernel, 2, sizeof(cl_mem), &d_value_sum_per_superpixel);
err |= clSetKernelArg(averageKernel, 3, sizeof(cl_mem), &d_pixel_counts_per_superpixel);
err |= clSetKernelArg(averageKernel, 4, localMemSize, nullptr); // Local memory for sums
err |= clSetKernelArg(averageKernel, 5, localCountSize, nullptr); // Local memory for counts
err |= clSetKernelArg(averageKernel, 6, sizeof(int), &numSuperpixels);
err |= clSetKernelArg(averageKernel, 7, sizeof(int), &numPixels);
// computeKernel
err |= clSetKernelArg(computeKernel, 0, sizeof(cl_mem), &d_value_sum_per_superpixel);
err |= clSetKernelArg(computeKernel, 1, sizeof(cl_mem), &d_pixel_counts_per_superpixel);
err |= clSetKernelArg(computeKernel, 2, sizeof(cl_mem), &d_averages);
err |= clSetKernelArg(computeKernel, 3, sizeof(int), &numSuperpixels);
// assignAvgKernel
err |= clSetKernelArg(assignAvgKernel, 0, sizeof(cl_mem), &d_labels);
err |= clSetKernelArg(assignAvgKernel, 1, sizeof(cl_mem), &d_averages);
err |= clSetKernelArg(assignAvgKernel, 2, sizeof(cl_mem), &d_avgImage1);
err |= clSetKernelArg(assignAvgKernel, 3, sizeof(int), &numPixels);
}
if (err != CL_SUCCESS) {
throw std::runtime_error("Failed to set kernel arguments");
}
}
// Release device buffers
void SLICSuperpixels::releaseBuffers() {
if (d_clusters) clReleaseMemObject(d_clusters);
if (d_accum) clReleaseMemObject(d_accum);
if (d_counts) clReleaseMemObject(d_counts);
// Release single map buffers
if (d_pixel_counts_per_superpixel) clReleaseMemObject(d_pixel_counts_per_superpixel);
if (d_value_sum_per_superpixel) clReleaseMemObject(d_value_sum_per_superpixel);
if (d_averages) clReleaseMemObject(d_averages);
// Release two-map buffers
if (d_value_sum_per_superpixel1) clReleaseMemObject(d_value_sum_per_superpixel1);
if (d_value_sum_per_superpixel2) clReleaseMemObject(d_value_sum_per_superpixel2);
if (d_pixel_counts_per_superpixel1) clReleaseMemObject(d_pixel_counts_per_superpixel1);
if (d_pixel_counts_per_superpixel2) clReleaseMemObject(d_pixel_counts_per_superpixel2);
if (d_averages1) clReleaseMemObject(d_averages1);
if (d_averages2) clReleaseMemObject(d_averages2);
d_clusters = d_accum = d_counts = nullptr;
d_pixel_counts_per_superpixel = d_value_sum_per_superpixel = d_averages = nullptr;
d_value_sum_per_superpixel1 = d_value_sum_per_superpixel2 = nullptr;
d_pixel_counts_per_superpixel1 = d_pixel_counts_per_superpixel2 = nullptr;
d_averages1 = d_averages2 = nullptr;
}
// Only run kernels and read back results, using external d_image
void SLICSuperpixels::processImage() {
cl_int err;
clFinish(queue);
// Execute initialization kernel
err = clEnqueueNDRangeKernel(queue, initKernel, 1, nullptr, &globalSize, nullptr, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to execute init kernel");
// assignPixels kernel
err = clEnqueueNDRangeKernel(queue, assignKernel, 2, nullptr, globalSize2, nullptr, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to execute assign kernel");
// accumulateClusters kernel
err = clEnqueueNDRangeKernel(queue, accumulateKernel, 1, nullptr, &globalSize3, nullptr, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to execute accumulate kernel");
// finalizeClusters kernel
size_t clusterGlobalSize = numSuperpixels;
err = clEnqueueNDRangeKernel(queue, finalizeKernel, 1, nullptr, &clusterGlobalSize, nullptr, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to execute finalize kernel");
// Launch accumulation kernel for averaging with appropriate work group size
size_t localWorkSize = 256; // Typical work group size
size_t globalWorkSize = ((numPixels + localWorkSize - 1) / localWorkSize) * localWorkSize;
err = clEnqueueNDRangeKernel(queue, averageKernel, 1, nullptr, &globalWorkSize, &localWorkSize, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to launch average kernel");
// Launch compute_averages kernel
size_t computeGlobalSize = numSuperpixels;
err = clEnqueueNDRangeKernel(queue, computeKernel, 1, nullptr, &computeGlobalSize, nullptr, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to launch compute_averages kernel");
// Launch assign_superpixel_average kernel
err = clEnqueueNDRangeKernel(queue, assignAvgKernel, 1, nullptr, &globalSize3, nullptr, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to launch assign_superpixel_average kernel");
}
// Only run kernels and read back results, using external d_image
void SLICSuperpixels::processTwoImages() {
cl_int err;
clFinish(queue);
// Execute initialization kernel
err = clEnqueueNDRangeKernel(queue, initKernel, 1, nullptr, &globalSize, nullptr, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to execute init kernel");
// assignPixels kernel
err = clEnqueueNDRangeKernel(queue, assignKernel, 2, nullptr, globalSize2, nullptr, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to execute assign kernel");
// accumulateClusters kernel
err = clEnqueueNDRangeKernel(queue, accumulateKernel, 1, nullptr, &globalSize3, nullptr, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to execute accumulate kernel");
// finalizeClusters kernel
size_t clusterGlobalSize = numSuperpixels;
err = clEnqueueNDRangeKernel(queue, finalizeKernel, 1, nullptr, &clusterGlobalSize, nullptr, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to execute finalize kernel");
// Launch accumulation kernel for averaging with appropriate work group size
size_t localWorkSize = 256; // Typical work group size
size_t globalWorkSize = ((numPixels + localWorkSize - 1) / localWorkSize) * localWorkSize;
err = clEnqueueNDRangeKernel(queue, averageKernel2, 1, nullptr, &globalWorkSize, &localWorkSize, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to launch average2 kernel");
// Launch compute_averages kernel
size_t computeGlobalSize = numSuperpixels;
err = clEnqueueNDRangeKernel(queue, computeKernel2, 1, nullptr, &computeGlobalSize, nullptr, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to launch compute_averages2 kernel");
// Launch assign_superpixel_average kernel
err = clEnqueueNDRangeKernel(queue, assignAvgKernel2, 1, nullptr, &globalSize3, nullptr, 0, nullptr, nullptr);
clFinish(queue);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to launch assign_superpixel_average2 kernel");
}
/*
int main(int argc, char** argv) {
try {
// 1. Load and prepare the image
cv::Mat image;
if (argc > 1) {
image = cv::imread(argv[1]);
}
if (image.empty()) {
std::cerr << "Failed to load image." << std::endl;
return -1;
}
int width = image.cols;
int height = image.rows;
int numSuperpixels = 1000;
float compactness = 36.0f;
int iterations = 1;
int regionSize = static_cast<int>(std::round(std::sqrt((width * height) / (float)numSuperpixels)));
if (regionSize < 1) regionSize = 1;
// 2. Create OpenCL context and command queue
cl_int err;
cl_platform_id platform;
cl_device_id device;
cl_context context;
cl_command_queue queue;
err = clGetPlatformIDs(1, &platform, nullptr);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to get OpenCL platform");
err = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, nullptr);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to get OpenCL device");
context = clCreateContext(nullptr, 1, &device, nullptr, nullptr, &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create OpenCL context");
queue = clCreateCommandQueue(context, device, 0, &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create OpenCL command queue");
// 3. Initialize image buffers on GPU
cv::Mat imageFloat, imagegray, opticalFlow;
image.convertTo(imageFloat, CV_32FC3, 1.0 / 255.0);
cv::cvtColor(image, imagegray, cv::COLOR_BGR2GRAY);
imagegray.convertTo(opticalFlow, CV_32FC1);
cl_mem d_image = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
width * height * 3 * sizeof(float), imageFloat.data, &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create image buffer");
cl_mem d_labels = clCreateBuffer(context, CL_MEM_READ_WRITE,
width * height * sizeof(int), nullptr, &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create labels buffer");
cl_mem d_opticalFlowMap1 = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
width * height * sizeof(float), opticalFlow.data, &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create optical flow map 1 buffer");
cl_mem d_opticalFlowMap2 = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
width * height * sizeof(float), opticalFlow.data, &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create optical flow map 2 buffer");
cl_mem d_avgImage1 = clCreateBuffer(context, CL_MEM_WRITE_ONLY, width * height * sizeof(float), nullptr, &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create output buffer 1");
cl_mem d_avgImage2 = clCreateBuffer(context, CL_MEM_WRITE_ONLY, width * height * sizeof(float), nullptr, &err);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to create output buffer 2");
// Zero initialize the output buffer
std::vector<float> zeroImage(width * height, 0.0f);
err = clEnqueueWriteBuffer(queue, d_avgImage1, CL_TRUE, 0, width * height * sizeof(float), zeroImage.data(), 0, nullptr, nullptr);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to initialize output buffer 1");
err = clEnqueueWriteBuffer(queue, d_avgImage2, CL_TRUE, 0, width * height * sizeof(float), zeroImage.data(), 0, nullptr, nullptr);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to initialize output buffer 2");
clFinish(queue);
auto start = std::chrono::high_resolution_clock::now();
clFinish(queue);
auto end = std::chrono::high_resolution_clock::now();
double cl_ms = static_cast<double>(std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count());
std::cout << "clFinish took " << cl_ms << " ms." << std::endl;
// 4. Create and configure SLIC object
bool processTwoMaps = true;
SLICSuperpixels slic(context, queue, device, compactness, numSuperpixels, iterations);
slic.allocateBuffers(width, height, processTwoMaps);
slic.setKernelArgs(d_image, d_labels, d_opticalFlowMap1, d_opticalFlowMap2, d_avgImage1, d_avgImage2);
// 5. Run SLIC
clFinish(queue);
auto slic_start = std::chrono::high_resolution_clock::now();
// slic.processImage(); // result saves in d_labels and d_avgImage1
slic.processTwoImages(); // result saves in d_labels and d_avgImage1 and d_avgImage2
clFinish(queue);
auto slic_end = std::chrono::high_resolution_clock::now();
double slic_ms = static_cast<double>(std::chrono::duration_cast<std::chrono::milliseconds>(slic_end - slic_start).count());
std::cout << "SLIC superpixel computation took " << slic_ms << " ms." << std::endl;
// Read back labels results
cv::Mat labels(height, width, CV_32S);
err = clEnqueueReadBuffer(queue, d_labels, CL_TRUE, 0, width * height * sizeof(int), labels.data, 0, nullptr, nullptr);
if (err != CL_SUCCESS) {
throw std::runtime_error("Failed to read labels buffer");
}
// 6. Visualize or save results as needed
cv::Mat result = image.clone();
for (int y = 1; y < image.rows - 1; y++) {
for (int x = 1; x < image.cols - 1; x++) {
int label = labels.at<int>(y, x);
if (labels.at<int>(y-1, x) != label || labels.at<int>(y+1, x) != label ||
labels.at<int>(y, x-1) != label || labels.at<int>(y, x+1) != label) {
result.at<cv::Vec3b>(y, x) = cv::Vec3b(0, 0, 255); // Red boundary
}
}
}
cv::imshow("Original Image", image);
cv::imshow("SLIC Superpixels", result);
cv::imwrite("original.jpg", image);
cv::imwrite("superpixels.jpg", result);
// Read back averaged optical flow results
cv::Mat avgImage1(height, width, CV_32FC1);
cv::Mat avgImage2(height, width, CV_32FC1);
err = clEnqueueReadBuffer(queue, d_avgImage1, CL_TRUE, 0, width * height * sizeof(float), avgImage1.data, 0, nullptr, nullptr);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to read output buffer 1");
err = clEnqueueReadBuffer(queue, d_avgImage2, CL_TRUE, 0, width * height * sizeof(float), avgImage2.data, 0, nullptr, nullptr);
if (err != CL_SUCCESS) throw std::runtime_error("Failed to read output buffer 2");
// Convert to 8-bit by rounding
cv::Mat avgImage8U1;
cv::Mat avgImage8U2;
avgImage1.convertTo(avgImage8U1, CV_8UC1);
avgImage2.convertTo(avgImage8U2, CV_8UC1);
// Show and save the 8-bit version
cv::imshow("Superpixel Averaged Optical Flow 1", avgImage8U1);
cv::imshow("Superpixel Averaged Optical Flow 2", avgImage8U2);
cv::imwrite("superpixel_averaged_optical_flow_1.jpg", avgImage8U1);
cv::imwrite("superpixel_averaged_optical_flow_2.jpg", avgImage8U2);
// // Run OpenCV's SLIC implementation for comparison
// cv::Mat labels_cv;
// cv::Mat result_cv = image.clone();
// cv::Mat avgImage_cv = cv::Mat::zeros(height, width, CV_32FC1);
// std::vector<cv::Mat> superpixel_means;
// cv::Mat avgImage8U_cv;
// auto cv_slic_start = std::chrono::high_resolution_clock::now();
// cv::Ptr<cv::ximgproc::SuperpixelSLIC> slic_cv = cv::ximgproc::createSuperpixelSLIC(image, cv::ximgproc::SLIC, regionSize, static_cast<float>(compactness*2.0));
// slic_cv->iterate(iterations);
// slic_cv->getLabels(labels_cv);
// auto cv_slic_end = std::chrono::high_resolution_clock::now();
// double cv_slic_ms = static_cast<double>(std::chrono::duration_cast<std::chrono::milliseconds>(cv_slic_end - cv_slic_start).count());
// std::cout << "OpenCV SLIC superpixel computation took " << cv_slic_ms << " ms." << std::endl;
// // Visualize OpenCV SLIC result
// for (int y = 1; y < image.rows - 1; y++) {
// for (int x = 1; x < image.cols - 1; x++) {
// int label = labels_cv.at<int>(y, x);
// if (labels_cv.at<int>(y-1, x) != label || labels_cv.at<int>(y+1, x) != label ||
// labels_cv.at<int>(y, x-1) != label || labels_cv.at<int>(y, x+1) != label) {
// result_cv.at<cv::Vec3b>(y, x) = cv::Vec3b(0, 0, 255); // Green boundary
// }
// }
// }
// cv::imshow("OpenCV SLIC Superpixels", result_cv);
// cv::imwrite("opencv_superpixels.jpg", result_cv);
// auto cv_avg_start = std::chrono::high_resolution_clock::now();