-
Notifications
You must be signed in to change notification settings - Fork 0
/
det_pp.cpp
451 lines (424 loc) · 18.8 KB
/
det_pp.cpp
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
#include "det_pp.h"
void box_nms(std::vector<BoxInfo>& input_boxes, float NMS_THRESH) {
std::sort(input_boxes.begin(), input_boxes.end(), [](BoxInfo a, BoxInfo b) { return a.score > b.score; });
std::vector<float> vArea(input_boxes.size());
for (int i = 0; i < int(input_boxes.size()); ++i) {
vArea[i] = (input_boxes.at(i).x2 - input_boxes.at(i).x1) * (input_boxes.at(i).y2 - input_boxes.at(i).y1);
}
for (int i = 0; i < int(input_boxes.size()); ++i) {
for (int j = i + 1; j < int(input_boxes.size());) {
float xx1 = std::max(input_boxes[i].x1, input_boxes[j].x1);
float yy1 = std::max(input_boxes[i].y1, input_boxes[j].y1);
float xx2 = std::min(input_boxes[i].x2, input_boxes[j].x2);
float yy2 = std::min(input_boxes[i].y2, input_boxes[j].y2);
float w = std::max(float(0), xx2 - xx1);
float h = std::max(float(0), yy2 - yy1);
float inter = w * h;
float ovr = inter / (vArea[i] + vArea[j] - inter);
if (ovr >= NMS_THRESH) {
input_boxes.erase(input_boxes.begin() + j);
vArea.erase(vArea.begin() + j);
} else {
j++;
}
}
}
}
void slot_nms(std::vector<SlotInfo>& input_slots, float xThr, float yThr) {
std::sort(input_slots.begin(), input_slots.end(), [](SlotInfo a, SlotInfo b) { return a.score > b.score; });
for (int i = 0; i < int(input_slots.size()); ++i) {
float cx1 = (input_slots[i].x1 + input_slots[i].x2) / 2;
float cy1 = (input_slots[i].y1 + input_slots[i].y2) / 2;
for (int j = i + 1; j < int(input_slots.size());) {
float cx2 = (input_slots[j].x1 + input_slots[j].x2) / 2;
float cy2 = (input_slots[j].y1 + input_slots[j].y2) / 2;
if ((abs(cx2 - cx1) < xThr) && (abs(cy2 - cy1) < yThr)) {
input_slots.erase(input_slots.begin() + j);
} else {
j++;
}
}
}
}
inline float sigmoid(float x) {
// return x;
return 1.0f / (1.0f + std::exp(-1.0f * x));
}
/*
for slot det(CHW)
*/
void get_yolo_slot_output(std::vector<cv::Mat> feat, int w, int h, int stride, int anchor_num, int class_num, float conf_thr, int yoloOutNum, int Input_w, int Input_h, float anchors[3][3][2], std::vector<SlotInfo>& slots) {
float* pO = nullptr;
int outSize = w * h;
int num_grid_x = w;
int num_grid_y = h;
float ratio = float(Input_w) / (w * stride);
//float ratioh=float(Input_h)/(h*stride);
float cx, cy, length, cos1, sin1, cos2, sin2, bp, cp, maxCP;
for (int i = 0; i < yoloOutNum; i++) {
//pO = &feat[0];
pO = (float*)feat[0].data;
for (int k = 0; k < anchor_num; k++) {
int xBeginIndex = k * (8 + class_num) * outSize;
int yBeginIndex = xBeginIndex + outSize;
int c1BeginIndex = yBeginIndex + outSize;
int s1BeginIndex = c1BeginIndex + outSize;
int c2BeginIndex = s1BeginIndex + outSize;
int s2BeginIndex = c2BeginIndex + outSize;
int lBeginIndex = s2BeginIndex + outSize;
int bpBeginIndex = lBeginIndex + outSize;
int cpBeginIndex = bpBeginIndex + outSize;
for (int y = 0; y < num_grid_y; y++) {
for (int x = 0; x < num_grid_x; x++) {
int index = y * num_grid_x + x;
int bpIndex = bpBeginIndex + index;
// float bp = sigmoid(pO[bpIndex]);
float bp = pO[bpIndex];
if (bp >= conf_thr) {
int classId = 0;
float maxCP = 0.0;
for (int o = 0; o < class_num; o++) {
// float cp = sigmoid(pO[cpBeginIndex + o * outSize + index]);
float cp = pO[cpBeginIndex + o * outSize + index];
if (cp > maxCP) {
maxCP = cp;
classId = o;
}
}
float score = maxCP * bp;
// fprintf(stderr,"slot score: %f\n",score);
if (score > conf_thr) {
cx = (2.0f * (pO[xBeginIndex + index]) - 0.5 + x) * stride;
cy = (2.0f * (pO[yBeginIndex + index]) - 0.5 + y) * stride;
cos1 = 2.0 * (pO[c1BeginIndex + index] - 0) - 1.0;
sin1 = 2.0 * (pO[s1BeginIndex + index] - 0) - 1.0;
cos2 = 2.0 * (pO[c2BeginIndex + index] - 0) - 1.0;
sin2 = 2.0 * (pO[s2BeginIndex + index] - 0) - 1.0;
length = (pO[lBeginIndex + index] - 0) * anchors[i][k][0];
float x1 = cx - length * cos1;
float y1 = cy - length * sin1;
float x2 = cx + length * cos1;
float y2 = cy + length * sin1;
x1 *= ratio;
y1 *= ratio;
x2 *= ratio;
y2 *= ratio;
cx *= ratio;
cy *= ratio;
length *= ratio;
slots.push_back({x1, y1, x2, y2, cos2, sin2, score, classId, cx, cy, length});
// fprintf(stderr,"slot coord %f, %f, %f, %f,%f, %f,%f,%d\n",cx, cy, cos1, sin1, cos2, sin2, score, classId);
if (slots.size() > 1000) {
return;
}
}
}
}
}
}
}
}
/**
for CHW OD out
*/
void get_yolo_output3(std::vector<float> feat, int w, int h, int yoloOut, int anchor_num, int class_num, int stride, float conf_thr, int Input_w, int Input_h, float anchor_yolov5n[3][3][2], std::vector<BoxInfo>& boxes) {
float* pO = nullptr;
pO = &feat[0];
int yoloOutSize = w * h;
int num_grid_x = w;
int num_grid_y = h;
float ratiow = float(Input_w) / (w * stride);
float ratioh = float(Input_h) / (h * stride);
for (int k = 0; k < anchor_num; k++) {
int xBeginIndex = k * (5 + class_num) * yoloOutSize;
int yBeginIndex = xBeginIndex + yoloOutSize;
int wBeginIndex = yBeginIndex + yoloOutSize;
int hBeginIndex = wBeginIndex + yoloOutSize;
int bpBeginIndex = hBeginIndex + yoloOutSize;
int cpBeginIndex = bpBeginIndex + yoloOutSize;
for (int y = 0; y < num_grid_y; y++) {
for (int x = 0; x < num_grid_x; x++) {
int index = y * num_grid_x + x;
int bpIndex = bpBeginIndex + index;
float bp = sigmoid(pO[bpIndex]);
if (bp >= conf_thr) {
int classId = 0;
float maxCP = 0.0;
for (int o = 0; o < class_num; o++) {
float cp = sigmoid(pO[cpBeginIndex + o * num_grid_x * num_grid_y + index]);
if (cp > maxCP) {
maxCP = cp;
classId = o;
}
}
// box.classIndex = classId;
float score = maxCP * bp;
if (score > conf_thr) {
// BBox box;
float bx = sigmoid(pO[xBeginIndex + index]);
float by = sigmoid(pO[yBeginIndex + index]);
float bw = sigmoid(pO[wBeginIndex + index]);
float bh = sigmoid(pO[hBeginIndex + index]);
bx = (2.0f * bx - 0.5 + x) * stride;
by = (2.0f * by - 0.5 + y) * stride;
bw = 2.0f * bw;
bw = bw * bw * anchor_yolov5n[yoloOut][k][0];
bh = 2.0f * bh;
bh = bh * bh * anchor_yolov5n[yoloOut][k][1];
// bx *= ratiow;
// by *= ratioh;
// bw *= ratiow;
// bh *= ratioh;
float left = bx - bw / 2;
float top = by - bh / 2;
float right = (left + bw);
float bottom = (top + bh);
boxes.push_back({left, top, right, bottom, score, classId});
// fprintf(stderr,"bxby %f, %f, %f, %f,%f, %d\n",left, top, right, bottom, score, classId);
if (boxes.size() > 1000) {
return;
}
}
}
}
}
}
}
void get_yolo_one_boxes(std::vector<float> feat, int class_num, float obj_thr, int Input_w, int Input_h, int img_w, int img_h, std::vector<BoxInfo>& boxes) {
float size = 3 * (80 * 80 + 40 * 40 + 20 * 20);
float ratiow = float(Input_w) / img_w;
float ratioh = float(Input_h) / img_h;
int padw = 0;
int padh = 0;
int offset = class_num + 5;
int npos = 0;
for (int r = 0; r < size; r++) {
float box_score = feat[npos + 4];
if (box_score >= obj_thr) {
int cls_id = 0;
float max_cls_prob = 0.0;
for (int cls = 5; cls < class_num + 5; cls++) {
float p = feat[npos + cls];
if (p > max_cls_prob) {
max_cls_prob = p;
cls_id = cls - 5;
}
}
float cx = (feat[npos + 0]) * ratiow;
float cy = (feat[npos + 1]) * ratioh;
float cw = feat[npos + 2] * ratiow;
float ch = feat[npos + 3] * ratioh;
// cv::Mat confs = out.row(r).colRange(5, 85);
// confs *= sc;
// double minV = 0, maxV = 0;
// double *minI = &minV;
// double *maxI = &maxV;
// cv::minMaxIdx(confs, minI, maxI);
float left = (cx - cw / 2.0f);
float top = (cy - ch / 2.0f);
float right = (cx + cw / 2.0f);
float bottom = (cy + ch / 2.0f);
float score = box_score * max_cls_prob;
if (left <= 0 || top <= 0 || right > Input_w || bottom > Input_h || left > right || top > bottom || score < 0.5) {
npos += offset;
continue;
}
boxes.push_back({left, top, right, bottom, score, cls_id});
}
npos += offset;
}
}
//for HWC out
void get_yolo_boxes(std::vector<float> feat, int w, int h, int anchor_num, int class_num, float obj_thr, float cls_thr, int Input_w, int Input_h, int img_w, int img_h, float anchors[], std::vector<BoxInfo>& boxes) {
float ratiow = float(Input_w) / img_w;
float ratioh = float(Input_h) / img_h;
int padw = 0;
int padh = 0;
int offset = class_num + 5;
int num_grid_x = w;
int num_grid_y = h;
//const float *pdata = &(feat[0]);
int npos = 0;
{
for (int k = 0; k < anchor_num; k++) {
for (int i = 0; i < num_grid_y; i++) {
for (int j = 0; j < num_grid_x; j++) {
float box_score = sigmoid(feat[npos + 4]);
if (box_score > obj_thr) {
int class_idx = 0;
float max_class_socre = 0;
for (int m = 0; m < class_num; m++) {
float cp = sigmoid(feat[npos + 5 + m]);
if (cp > max_class_socre) {
max_class_socre = cp;
class_idx = m;
}
}
max_class_socre *= box_score;
if (max_class_socre > cls_thr) {
float cx = (2.0 * sigmoid(feat[npos]) + j - 0.5) * (img_w / w); /// cx
float cy = (2.0 * sigmoid(feat[npos + 1]) + i - 0.5) * (img_h / h); /// cy
float cw = anchors[2 * k] * std::pow(sigmoid(feat[npos + 2]) * 2, 2);
float ch = anchors[2 * k + 1] * std::pow(sigmoid(feat[npos + 3]) * 2, 2);
float bx = (cx - padw) * ratiow;
float by = (cy - padh) * ratioh;
float bw = cw * ratiow;
float bh = ch * ratioh;
float left = (bx - bw / 2.0f);
float top = (by - bh / 2.0f);
float right = (bx + bw / 2.0f);
float bottom = (by + bh / 2.0f);
// fprintf(stderr,"bxby %f, %f, %f, %f,%f, %d\n",left, top, right, bottom, max_class_socre, class_idx);
if (left <= 0 || top <= 0 || right > Input_w || bottom > Input_h) {
npos += offset;
continue;
}
if (boxes.size() < 1000) {
boxes.push_back({left, top, right, bottom, max_class_socre, class_idx});
}
}
}
//row_ind++;
npos += offset;
}
}
}
}
}
//void draw_box(cv::Mat & cv_mat, std::vector<BoxInfo> &boxes, const std::vector<std::string> &labels, int offset)
//{
// //int CNUM = labels.size();
// int CNUM = 2;
// // fprintf(stderr,"CNUM: %d\n",CNUM);
// cv::RNG rng(0xFFFFFFFF);
// cv::Scalar_<int> randColor[CNUM];
// for (int i = 0; i < CNUM; i++)
// rng.fill(randColor[i], cv::RNG::UNIFORM, 0, 256);
// int img_w = cv_mat.cols;
// int img_h = cv_mat.rows;
// for(auto box : boxes)
// {
// // int width = (box.x2-box.x1)*img_w;
// // int height = (box.y2-box.y1)*img_h;
// // int x = box.x1*img_w;
// // int y = box.y1*img_h;
// int width = (box.x2-box.x1);
// int height = (box.y2-box.y1);
// int x = box.x1;
// int y = box.y1;
// // fprintf(stderr,"box type: %s, confidence: %0.3f, axis: top-left-x: %d, top-left-y: %d, width: %d, height: %d\n",
// // labels[box.label+offset].c_str(), box.score, x, y, width, height);
// cv::Point p = cv::Point(x, y);
// cv::Rect rect = cv::Rect(x, y, width, height);
// cv::rectangle(cv_mat, rect, cv::Scalar(255,0,0));/*randColor[box.label+offset]);;*/
// std::string text = labels[box.label+offset] + ":" + std::to_string(box.score) ;
// cv::putText(cv_mat, text, p, cv::FONT_HERSHEY_PLAIN, 1, randColor[box.label+offset]);
// }
//}
// int astride = class_num + 5;
// int c_stride = 1;
// int j_stride = astride*anchor_num;
// int i_stride = j_stride*w;
// if (type == "CHW")
// {
// j_stride = 1;
// i_stride = w;
// c_stride = w*h;
// }
// for (int i = 0; i < h; i++)
// {
// for(int j = 0; j < w; j++)
// {
// for(int k = 0; k < anchor_num; k ++){
// float tx = feat[i*i_stride + j*j_stride + (0 + astride * k)*c_stride];
// float ty = feat[i*i_stride + j*j_stride + (1 + astride * k)*c_stride];
// float tw = feat[i*i_stride + j*j_stride + (2 + astride * k)*c_stride];
// float th = feat[i*i_stride + j*j_stride + (3 + astride * k)*c_stride];
// float cf = feat[i*i_stride + j*j_stride + (4 + astride * k)*c_stride];
// // float bx = (2.0*tx -0.5 + j) * (img_w / w);
// // float by = (2.0*ty -0.5 +i) * (img_h / h);
// float bx = (2.0*sigmoid(tx) -0.5 + j) * (img_w / w);
// float by = (2.0*sigmoid(ty) -0.5 +i) * (img_h / h);
// float bw = anchors[2*k] * std::pow(sigmoid(tw)*2,2) ;
// float bh = anchors[2*k+1] * std::pow(sigmoid(th)*2,2) ;
// // float bx = (sigmoid(tx) + j) / (float)w;
// // float by = (sigmoid(ty) + i) / (float)h;
// // float bw = anchors[2*k] * std::exp(tw) / (float)img_w;
// // float bh = anchors[2*k+1] * std::exp(th) / (float)img_h;
// float b_confidence = sigmoid(cf);
// // float b_confidence = cf;
// float b_class_score = 0.0f;
// int b_class_index = -1;
// for(int c = 0; c < class_num; c++)
// {
// float b_scores = b_confidence*sigmoid(feat[i*i_stride + j*j_stride + (5 + astride * k+c)*c_stride]);
// // float b_scores = b_confidence*(feat[i*i_stride + j*j_stride + (5 + astride * k+c)*c_stride]);
// if (b_scores > conf_thr && b_scores > b_class_score)
// {
// b_class_score = b_scores;
// b_class_index = c;
// }
// }
// if (b_class_index >= 0)
// {
// fprintf(stderr,"bxby %f, %f, %f, %f, %d\n",bx,by,bw,bh, b_class_index);
// boxes.push_back({(bx - bw /2.0f), (by - bh /2.0f), (bx + bw /2.0f), (by + bh /2.0f), b_class_score, b_class_index});
// }
// }
// }
// }
// void get_yolo_box2(std::vector<float> ort_outputs, int w, int h, int stride, int img_w, int img_h, float anchors[], float obj_thr, float score_thr)
// {
// int q,i,j,k;
// int newh = 640, neww = 640, padh = 0, padw = 0;
// float ratioh = (float)img_h / newh, ratiow = (float)img_w / neww;
// int classNum = 2;
// {
// const float* pdata = &(ort_outputs[0]);
// int offset = 2 + 5;
// int num_grid_x = w;
// int num_grid_y = h;
// for (q = 0; q < 3; q++) ///anchor
// {
// const float anchor_w = anchors[q * 2];
// const float anchor_h = anchors[q * 2 + 1];
// for (i = 0; i < num_grid_y; i++)
// {
// for (j = 0; j < num_grid_x; j++)
// {
// float box_score = sigmoid(pdata[4]);
// if (box_score > obj_thr)
// {
// int class_idx = 0;
// float max_class_socre = 0;
// for (k = 0; k < classNum; k++)
// {
// float cp = sigmoid(pdata[5 + k]);
// if (cp > max_class_socre)
// {
// max_class_socre = cp;
// class_idx = k;
// }
// }
// max_class_socre *= box_score;
// if (max_class_socre > score_thr)
// {
// float cx = (sigmoid(pdata[0]) + j) * stride; ///cx
// float cy = (sigmoid(pdata[1]) + i) * stride; ///cy
// float w = expf(pdata[2]) * anchor_w; ///w
// float h = expf(pdata[3]) * anchor_h; ///h
// cx = (cx - padw) * ratiow;
// cy = (cy - padh) * ratioh;
// w *= ratiow;
// h *= ratioh;
// // BBox boxInfo = BBox(class_idx, max_class_socre, cx - w / 2, cy - h / 2, w, h, 0);
// // generate_boxes.push_back(boxInfo);
// }
// }
// //row_ind++;
// pdata += (2 + 5);
// }
// }
// }
// }
// }