forked from hybridgroup/gocv
-
Notifications
You must be signed in to change notification settings - Fork 0
/
dnn.go
593 lines (512 loc) · 17.8 KB
/
dnn.go
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
package gocv
/*
#include <stdlib.h>
#include "dnn.h"
*/
import "C"
import (
"image"
"reflect"
"unsafe"
)
// Net allows you to create and manipulate comprehensive artificial neural networks.
//
// For further details, please see:
// https://docs.opencv.org/master/db/d30/classcv_1_1dnn_1_1Net.html
type Net struct {
// C.Net
p unsafe.Pointer
}
// NetBackendType is the type for the various different kinds of DNN backends.
type NetBackendType int
const (
// NetBackendDefault is the default backend.
NetBackendDefault NetBackendType = 0
// NetBackendHalide is the Halide backend.
NetBackendHalide NetBackendType = 1
// NetBackendOpenVINO is the OpenVINO backend.
NetBackendOpenVINO NetBackendType = 2
// NetBackendOpenCV is the OpenCV backend.
NetBackendOpenCV NetBackendType = 3
// NetBackendVKCOM is the Vulkan backend.
NetBackendVKCOM NetBackendType = 4
// NetBackendCUDA is the Cuda backend.
NetBackendCUDA NetBackendType = 5
)
// ParseNetBackend returns a valid NetBackendType given a string. Valid values are:
// - halide
// - openvino
// - opencv
// - vulkan
// - cuda
// - default
func ParseNetBackend(backend string) NetBackendType {
switch backend {
case "halide":
return NetBackendHalide
case "openvino":
return NetBackendOpenVINO
case "opencv":
return NetBackendOpenCV
case "vulkan":
return NetBackendVKCOM
case "cuda":
return NetBackendCUDA
default:
return NetBackendDefault
}
}
// NetTargetType is the type for the various different kinds of DNN device targets.
type NetTargetType int
const (
// NetTargetCPU is the default CPU device target.
NetTargetCPU NetTargetType = 0
// NetTargetFP32 is the 32-bit OpenCL target.
NetTargetFP32 NetTargetType = 1
// NetTargetFP16 is the 16-bit OpenCL target.
NetTargetFP16 NetTargetType = 2
// NetTargetVPU is the Movidius VPU target.
NetTargetVPU NetTargetType = 3
// NetTargetVulkan is the NVIDIA Vulkan target.
NetTargetVulkan NetTargetType = 4
// NetTargetFPGA is the FPGA target.
NetTargetFPGA NetTargetType = 5
// NetTargetCUDA is the CUDA target.
NetTargetCUDA NetTargetType = 6
// NetTargetCUDAFP16 is the CUDA target.
NetTargetCUDAFP16 NetTargetType = 7
)
// ParseNetTarget returns a valid NetTargetType given a string. Valid values are:
// - cpu
// - fp32
// - fp16
// - vpu
// - vulkan
// - fpga
// - cuda
// - cudafp16
func ParseNetTarget(target string) NetTargetType {
switch target {
case "cpu":
return NetTargetCPU
case "fp32":
return NetTargetFP32
case "fp16":
return NetTargetFP16
case "vpu":
return NetTargetVPU
case "vulkan":
return NetTargetVulkan
case "fpga":
return NetTargetFPGA
case "cuda":
return NetTargetCUDA
case "cudafp16":
return NetTargetCUDAFP16
default:
return NetTargetCPU
}
}
// Close Net
func (net *Net) Close() error {
C.Net_Close((C.Net)(net.p))
net.p = nil
return nil
}
// Empty returns true if there are no layers in the network.
//
// For further details, please see:
// https://docs.opencv.org/master/db/d30/classcv_1_1dnn_1_1Net.html#a6a5778787d5b8770deab5eda6968e66c
func (net *Net) Empty() bool {
return bool(C.Net_Empty((C.Net)(net.p)))
}
// SetInput sets the new value for the layer output blob.
//
// For further details, please see:
// https://docs.opencv.org/trunk/db/d30/classcv_1_1dnn_1_1Net.html#a672a08ae76444d75d05d7bfea3e4a328
func (net *Net) SetInput(blob Mat, name string) {
cName := C.CString(name)
defer C.free(unsafe.Pointer(cName))
C.Net_SetInput((C.Net)(net.p), blob.p, cName)
}
// Forward runs forward pass to compute output of layer with name outputName.
//
// For further details, please see:
// https://docs.opencv.org/trunk/db/d30/classcv_1_1dnn_1_1Net.html#a98ed94cb6ef7063d3697259566da310b
func (net *Net) Forward(outputName string) Mat {
cName := C.CString(outputName)
defer C.free(unsafe.Pointer(cName))
return newMat(C.Net_Forward((C.Net)(net.p), cName))
}
// ForwardLayers forward pass to compute outputs of layers listed in outBlobNames.
//
// For further details, please see:
// https://docs.opencv.org/3.4.1/db/d30/classcv_1_1dnn_1_1Net.html#adb34d7650e555264c7da3b47d967311b
func (net *Net) ForwardLayers(outBlobNames []string) (blobs []Mat) {
cMats := C.struct_Mats{}
C.Net_ForwardLayers((C.Net)(net.p), &(cMats), toCStrings(outBlobNames))
blobs = make([]Mat, cMats.length)
for i := C.int(0); i < cMats.length; i++ {
blobs[i].p = C.Mats_get(cMats, i)
addMatToProfile(blobs[i].p)
}
return
}
// SetPreferableBackend ask network to use specific computation backend.
//
// For further details, please see:
// https://docs.opencv.org/3.4/db/d30/classcv_1_1dnn_1_1Net.html#a7f767df11386d39374db49cd8df8f59e
func (net *Net) SetPreferableBackend(backend NetBackendType) error {
C.Net_SetPreferableBackend((C.Net)(net.p), C.int(backend))
return nil
}
// SetPreferableTarget ask network to make computations on specific target device.
//
// For further details, please see:
// https://docs.opencv.org/3.4/db/d30/classcv_1_1dnn_1_1Net.html#a9dddbefbc7f3defbe3eeb5dc3d3483f4
func (net *Net) SetPreferableTarget(target NetTargetType) error {
C.Net_SetPreferableTarget((C.Net)(net.p), C.int(target))
return nil
}
// ReadNet reads a deep learning network represented in one of the supported formats.
//
// For further details, please see:
// https://docs.opencv.org/3.4/d6/d0f/group__dnn.html#ga3b34fe7a29494a6a4295c169a7d32422
func ReadNet(model string, config string) Net {
cModel := C.CString(model)
defer C.free(unsafe.Pointer(cModel))
cConfig := C.CString(config)
defer C.free(unsafe.Pointer(cConfig))
return Net{p: unsafe.Pointer(C.Net_ReadNet(cModel, cConfig))}
}
// ReadNetBytes reads a deep learning network represented in one of the supported formats.
//
// For further details, please see:
// https://docs.opencv.org/master/d6/d0f/group__dnn.html#ga138439da76f26266fdefec9723f6c5cd
func ReadNetBytes(framework string, model []byte, config []byte) (Net, error) {
cFramework := C.CString(framework)
defer C.free(unsafe.Pointer(cFramework))
bModel, err := toByteArray(model)
if err != nil {
return Net{}, err
}
bConfig, err := toByteArray(config)
if err != nil {
return Net{}, err
}
return Net{p: unsafe.Pointer(C.Net_ReadNetBytes(cFramework, *bModel, *bConfig))}, nil
}
// ReadNetFromCaffe reads a network model stored in Caffe framework's format.
//
// For further details, please see:
// https://docs.opencv.org/master/d6/d0f/group__dnn.html#ga29d0ea5e52b1d1a6c2681e3f7d68473a
func ReadNetFromCaffe(prototxt string, caffeModel string) Net {
cprototxt := C.CString(prototxt)
defer C.free(unsafe.Pointer(cprototxt))
cmodel := C.CString(caffeModel)
defer C.free(unsafe.Pointer(cmodel))
return Net{p: unsafe.Pointer(C.Net_ReadNetFromCaffe(cprototxt, cmodel))}
}
// ReadNetFromCaffeBytes reads a network model stored in Caffe model in memory.
//
// For further details, please see:
// https://docs.opencv.org/master/d6/d0f/group__dnn.html#ga946b342af1355185a7107640f868b64a
func ReadNetFromCaffeBytes(prototxt []byte, caffeModel []byte) (Net, error) {
bPrototxt, err := toByteArray(prototxt)
if err != nil {
return Net{}, err
}
bCaffeModel, err := toByteArray(caffeModel)
if err != nil {
return Net{}, err
}
return Net{p: unsafe.Pointer(C.Net_ReadNetFromCaffeBytes(*bPrototxt, *bCaffeModel))}, nil
}
// ReadNetFromTensorflow reads a network model stored in Tensorflow framework's format.
//
// For further details, please see:
// https://docs.opencv.org/master/d6/d0f/group__dnn.html#gad820b280978d06773234ba6841e77e8d
func ReadNetFromTensorflow(model string) Net {
cmodel := C.CString(model)
defer C.free(unsafe.Pointer(cmodel))
return Net{p: unsafe.Pointer(C.Net_ReadNetFromTensorflow(cmodel))}
}
// ReadNetFromTensorflowBytes reads a network model stored in Tensorflow framework's format.
//
// For further details, please see:
// https://docs.opencv.org/master/d6/d0f/group__dnn.html#gacdba30a7c20db2788efbf5bb16a7884d
func ReadNetFromTensorflowBytes(model []byte) (Net, error) {
bModel, err := toByteArray(model)
if err != nil {
return Net{}, err
}
return Net{p: unsafe.Pointer(C.Net_ReadNetFromTensorflowBytes(*bModel))}, nil
}
// ReadNetFromTorch reads a network model stored in Torch framework's format (t7).
//
// check net.Empty() for read failure
//
// For further details, please see:
// https://docs.opencv.org/master/d6/d0f/group__dnn.html#gaaaed8c8530e9e92fe6647700c13d961e
func ReadNetFromTorch(model string) Net {
cmodel := C.CString(model)
defer C.free(unsafe.Pointer(cmodel))
return Net{p: unsafe.Pointer(C.Net_ReadNetFromTorch(cmodel))}
}
// ReadNetFromONNX reads a network model stored in ONNX framework's format.
//
// check net.Empty() for read failure
//
// For further details, please see:
// https://docs.opencv.org/master/d6/d0f/group__dnn.html#ga7faea56041d10c71dbbd6746ca854197
func ReadNetFromONNX(model string) Net {
cmodel := C.CString(model)
defer C.free(unsafe.Pointer(cmodel))
return Net{p: unsafe.Pointer(C.Net_ReadNetFromONNX(cmodel))}
}
// ReadNetFromONNXBytes reads a network model stored in ONNX framework's format.
//
// For further details, please see:
// https://docs.opencv.org/master/d6/d0f/group__dnn.html#ga9198ecaac7c32ddf0aa7a1bcbd359567
func ReadNetFromONNXBytes(model []byte) (Net, error) {
bModel, err := toByteArray(model)
if err != nil {
return Net{}, err
}
return Net{p: unsafe.Pointer(C.Net_ReadNetFromONNXBytes(*bModel))}, nil
}
// BlobFromImage creates 4-dimensional blob from image. Optionally resizes and crops
// image from center, subtract mean values, scales values by scalefactor,
// swap Blue and Red channels.
//
// For further details, please see:
// https://docs.opencv.org/trunk/d6/d0f/group__dnn.html#ga152367f253c81b53fe6862b299f5c5cd
func BlobFromImage(img Mat, scaleFactor float64, size image.Point, mean Scalar,
swapRB bool, crop bool) Mat {
sz := C.struct_Size{
width: C.int(size.X),
height: C.int(size.Y),
}
sMean := C.struct_Scalar{
val1: C.double(mean.Val1),
val2: C.double(mean.Val2),
val3: C.double(mean.Val3),
val4: C.double(mean.Val4),
}
return newMat(C.Net_BlobFromImage(img.p, C.double(scaleFactor), sz, sMean, C.bool(swapRB), C.bool(crop)))
}
// BlobFromImages Creates 4-dimensional blob from series of images.
// Optionally resizes and crops images from center, subtract mean values,
// scales values by scalefactor, swap Blue and Red channels.
//
// For further details, please see:
// https://docs.opencv.org/master/d6/d0f/group__dnn.html#ga2b89ed84432e4395f5a1412c2926293c
func BlobFromImages(imgs []Mat, blob *Mat, scaleFactor float64, size image.Point, mean Scalar,
swapRB bool, crop bool, ddepth MatType) {
cMatArray := make([]C.Mat, len(imgs))
for i, r := range imgs {
cMatArray[i] = r.p
}
cMats := C.struct_Mats{
mats: (*C.Mat)(&cMatArray[0]),
length: C.int(len(imgs)),
}
sz := C.struct_Size{
width: C.int(size.X),
height: C.int(size.Y),
}
sMean := C.struct_Scalar{
val1: C.double(mean.Val1),
val2: C.double(mean.Val2),
val3: C.double(mean.Val3),
val4: C.double(mean.Val4),
}
C.Net_BlobFromImages(cMats, blob.p, C.double(scaleFactor), sz, sMean, C.bool(swapRB), C.bool(crop), C.int(ddepth))
}
// ImagesFromBlob Parse a 4D blob and output the images it contains as
// 2D arrays through a simpler data structure (std::vector<cv::Mat>).
//
// For further details, please see:
// https://docs.opencv.org/master/d6/d0f/group__dnn.html#ga4051b5fa2ed5f54b76c059a8625df9f5
func ImagesFromBlob(blob Mat, imgs []Mat) {
cMats := C.struct_Mats{}
C.Net_ImagesFromBlob(blob.p, &(cMats))
// mv = make([]Mat, cMats.length)
for i := C.int(0); i < cMats.length; i++ {
imgs[i].p = C.Mats_get(cMats, i)
}
}
// GetBlobChannel extracts a single (2d)channel from a 4 dimensional blob structure
// (this might e.g. contain the results of a SSD or YOLO detection,
//
// a bones structure from pose detection, or a color plane from Colorization)
func GetBlobChannel(blob Mat, imgidx int, chnidx int) Mat {
return newMat(C.Net_GetBlobChannel(blob.p, C.int(imgidx), C.int(chnidx)))
}
// GetBlobSize retrieves the 4 dimensional size information in (N,C,H,W) order
func GetBlobSize(blob Mat) Scalar {
s := C.Net_GetBlobSize(blob.p)
return NewScalar(float64(s.val1), float64(s.val2), float64(s.val3), float64(s.val4))
}
// Layer is a wrapper around the cv::dnn::Layer algorithm.
type Layer struct {
// C.Layer
p unsafe.Pointer
}
// GetLayer returns pointer to layer with specified id from the network.
//
// For further details, please see:
// https://docs.opencv.org/master/db/d30/classcv_1_1dnn_1_1Net.html#a70aec7f768f38c32b1ee25f3a56526df
func (net *Net) GetLayer(layer int) Layer {
return Layer{p: unsafe.Pointer(C.Net_GetLayer((C.Net)(net.p), C.int(layer)))}
}
// GetPerfProfile returns overall time for inference and timings (in ticks) for layers
//
// For further details, please see:
// https://docs.opencv.org/master/db/d30/classcv_1_1dnn_1_1Net.html#a06ce946f675f75d1c020c5ddbc78aedc
func (net *Net) GetPerfProfile() float64 {
return float64(C.Net_GetPerfProfile((C.Net)(net.p)))
}
// GetUnconnectedOutLayers returns indexes of layers with unconnected outputs.
//
// For further details, please see:
// https://docs.opencv.org/master/db/d30/classcv_1_1dnn_1_1Net.html#ae62a73984f62c49fd3e8e689405b056a
func (net *Net) GetUnconnectedOutLayers() (ids []int) {
cids := C.IntVector{}
C.Net_GetUnconnectedOutLayers((C.Net)(net.p), &cids)
defer C.free(unsafe.Pointer(cids.val))
h := &reflect.SliceHeader{
Data: uintptr(unsafe.Pointer(cids.val)),
Len: int(cids.length),
Cap: int(cids.length),
}
pcids := *(*[]C.int)(unsafe.Pointer(h))
for i := 0; i < int(cids.length); i++ {
ids = append(ids, int(pcids[i]))
}
return
}
// GetLayerNames returns all layer names.
//
// For furtherdetails, please see:
// https://docs.opencv.org/master/db/d30/classcv_1_1dnn_1_1Net.html#ae8be9806024a0d1d41aba687cce99e6b
func (net *Net) GetLayerNames() (names []string) {
cstrs := C.CStrings{}
defer C.CStrings_Close(cstrs)
C.Net_GetLayerNames((C.Net)(net.p), &cstrs)
return toGoStrings(cstrs)
}
// Close Layer
func (l *Layer) Close() error {
C.Layer_Close((C.Layer)(l.p))
l.p = nil
return nil
}
// GetName returns name for this layer.
func (l *Layer) GetName() string {
return C.GoString(C.Layer_GetName((C.Layer)(l.p)))
}
// GetType returns type for this layer.
func (l *Layer) GetType() string {
return C.GoString(C.Layer_GetType((C.Layer)(l.p)))
}
// InputNameToIndex returns index of input blob in input array.
//
// For further details, please see:
// https://docs.opencv.org/master/d3/d6c/classcv_1_1dnn_1_1Layer.html#a60ffc8238f3fa26cd3f49daa7ac0884b
func (l *Layer) InputNameToIndex(name string) int {
cName := C.CString(name)
defer C.free(unsafe.Pointer(cName))
return int(C.Layer_InputNameToIndex((C.Layer)(l.p), cName))
}
// OutputNameToIndex returns index of output blob in output array.
//
// For further details, please see:
// https://docs.opencv.org/master/d3/d6c/classcv_1_1dnn_1_1Layer.html#a60ffc8238f3fa26cd3f49daa7ac0884b
func (l *Layer) OutputNameToIndex(name string) int {
cName := C.CString(name)
defer C.free(unsafe.Pointer(cName))
return int(C.Layer_OutputNameToIndex((C.Layer)(l.p), cName))
}
// NMSBoxes performs non maximum suppression given boxes and corresponding scores.
//
// For futher details, please see:
// https://docs.opencv.org/4.4.0/d6/d0f/group__dnn.html#ga9d118d70a1659af729d01b10233213ee
func NMSBoxes(bboxes []image.Rectangle, scores []float32, scoreThreshold float32, nmsThreshold float32) (indices []int) {
bboxesRectArr := []C.struct_Rect{}
for _, v := range bboxes {
bbox := C.struct_Rect{
x: C.int(v.Min.X),
y: C.int(v.Min.Y),
width: C.int(v.Size().X),
height: C.int(v.Size().Y),
}
bboxesRectArr = append(bboxesRectArr, bbox)
}
bboxesRects := C.Rects{
rects: (*C.Rect)(&bboxesRectArr[0]),
length: C.int(len(bboxes)),
}
scoresFloats := []C.float{}
for _, v := range scores {
scoresFloats = append(scoresFloats, C.float(v))
}
scoresVector := C.struct_FloatVector{}
scoresVector.val = (*C.float)(&scoresFloats[0])
scoresVector.length = (C.int)(len(scoresFloats))
indicesVector := C.IntVector{}
C.NMSBoxes(bboxesRects, scoresVector, C.float(scoreThreshold), C.float(nmsThreshold), &indicesVector)
defer C.free(unsafe.Pointer(indicesVector.val))
h := &reflect.SliceHeader{
Data: uintptr(unsafe.Pointer(indicesVector.val)),
Len: int(indicesVector.length),
Cap: int(indicesVector.length),
}
ptr := *(*[]C.int)(unsafe.Pointer(h))
indices = make([]int, indicesVector.length)
for i := 0; i < int(indicesVector.length); i++ {
indices[i] = int(ptr[i])
}
return
}
// NMSBoxesWithParams performs non maximum suppression given boxes and corresponding scores.
//
// For futher details, please see:
// https://docs.opencv.org/4.4.0/d6/d0f/group__dnn.html#ga9d118d70a1659af729d01b10233213ee
func NMSBoxesWithParams(bboxes []image.Rectangle, scores []float32, scoreThreshold float32, nmsThreshold float32, eta float32, topK int) (indices []int) {
bboxesRectArr := []C.struct_Rect{}
for _, v := range bboxes {
bbox := C.struct_Rect{
x: C.int(v.Min.X),
y: C.int(v.Min.Y),
width: C.int(v.Size().X),
height: C.int(v.Size().Y),
}
bboxesRectArr = append(bboxesRectArr, bbox)
}
bboxesRects := C.Rects{
rects: (*C.Rect)(&bboxesRectArr[0]),
length: C.int(len(bboxes)),
}
scoresFloats := []C.float{}
for _, v := range scores {
scoresFloats = append(scoresFloats, C.float(v))
}
scoresVector := C.struct_FloatVector{}
scoresVector.val = (*C.float)(&scoresFloats[0])
scoresVector.length = (C.int)(len(scoresFloats))
indicesVector := C.IntVector{}
C.NMSBoxesWithParams(bboxesRects, scoresVector, C.float(scoreThreshold), C.float(nmsThreshold), &indicesVector, C.float(eta), C.int(topK))
defer C.free(unsafe.Pointer(indicesVector.val))
h := &reflect.SliceHeader{
Data: uintptr(unsafe.Pointer(indicesVector.val)),
Len: int(indicesVector.length),
Cap: int(indicesVector.length),
}
ptr := *(*[]C.int)(unsafe.Pointer(h))
indices = make([]int, indicesVector.length)
for i := 0; i < int(indicesVector.length); i++ {
indices[i] = int(ptr[i])
}
return
}