forked from shirokunet/MobileNet-SSD-TensorRT-Qt
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.cpp
executable file
·166 lines (139 loc) · 4.91 KB
/
main.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
#include "common.h"
#include "cudaUtility.h"
#include "mathFunctions.h"
#include "pluginImplement.h"
#include "tensorNet.h"
#include "loadImage.h"
#include <chrono>
const char* model = "../../model/MobileNetSSD_deploy_iplugin.prototxt";
const char* weight = "../../model/MobileNetSSD_deploy.caffemodel";
const char* INPUT_BLOB_NAME = "data";
const char* OUTPUT_BLOB_NAME = "detection_out";
static const uint32_t BATCH_SIZE = 1;
class Timer {
public:
void tic() {
start_ticking_ = true;
start_ = std::chrono::high_resolution_clock::now();
}
void toc() {
if (!start_ticking_)return;
end_ = std::chrono::high_resolution_clock::now();
start_ticking_ = false;
t = std::chrono::duration<double, std::milli>(end_ - start_).count();
std::cout << "Time: " << t << " ms" << std::endl;
}
double t;
private:
bool start_ticking_ = false;
std::chrono::time_point<std::chrono::high_resolution_clock> start_;
std::chrono::time_point<std::chrono::high_resolution_clock> end_;
};
/* *
* @TODO: unifiedMemory is used here under -> ( cudaMallocManaged )
* */
float* allocateMemory(DimsCHW dims, char* info)
{
float* ptr;
size_t size;
std::cout << "Allocate memory: " << info << std::endl;
size = BATCH_SIZE * dims.c() * dims.h() * dims.w();
assert(!cudaMallocManaged( &ptr, size*sizeof(float)));
return ptr;
}
void loadImg( cv::Mat &input, int re_width, int re_height, float *data_unifrom,const float3 mean,const float scale )
{
int i;
int j;
int line_offset;
int offset_g;
int offset_r;
cv::Mat dst;
unsigned char *line = NULL;
float *unifrom_data = data_unifrom;
cv::resize( input, dst, cv::Size( re_width, re_height ), (0.0), (0.0), cv::INTER_LINEAR );
offset_g = re_width * re_height;
offset_r = re_width * re_height * 2;
for( i = 0; i < re_height; ++i )
{
line = dst.ptr< unsigned char >( i );
line_offset = i * re_width;
for( j = 0; j < re_width; ++j )
{
// b
unifrom_data[ line_offset + j ] = (( float )(line[ j * 3 ] - mean.x) * scale);
// g
unifrom_data[ offset_g + line_offset + j ] = (( float )(line[ j * 3 + 1 ] - mean.y) * scale);
// r
unifrom_data[ offset_r + line_offset + j ] = (( float )(line[ j * 3 + 2 ] - mean.z) * scale);
}
}
}
int main(int argc, char *argv[])
{
std::vector<std::string> output_vector = {OUTPUT_BLOB_NAME};
TensorNet tensorNet;
tensorNet.LoadNetwork(model,weight,INPUT_BLOB_NAME, output_vector,BATCH_SIZE);
DimsCHW dimsData = tensorNet.getTensorDims(INPUT_BLOB_NAME);
DimsCHW dimsOut = tensorNet.getTensorDims(OUTPUT_BLOB_NAME);
float* data = allocateMemory( dimsData , (char*)"input blob");
std::cout << "allocate data" << std::endl;
float* output = allocateMemory( dimsOut , (char*)"output blob");
std::cout << "allocate output" << std::endl;
int height = 300;
int width = 300;
cv::Mat frame,srcImg;
void* imgCPU;
void* imgCUDA;
Timer timer;
std::string imgFile = "../../testPic/test.jpg";
frame = cv::imread(imgFile);
if (frame.empty())
{
cout <<"Could not find the image."<<endl;
return 0;
}
srcImg = frame.clone();
cv::resize(frame, frame, cv::Size(height,width));
const size_t size = width * height * sizeof(float3);
if( CUDA_FAILED( cudaMalloc( &imgCUDA, size)) )
{
cout <<"Cuda Memory allocation error occured."<<endl;
return 0;
}
void* imgData = malloc(size);
memset(imgData,0,size);
timer.tic();
loadImg(frame,height,width,(float*)imgData,make_float3(127.5,127.5,127.5),0.007843);
cudaMemcpyAsync(imgCUDA,imgData,size,cudaMemcpyHostToDevice);
void* buffers[] = { imgCUDA, output };
tensorNet.imageInference( buffers, output_vector.size() + 1, BATCH_SIZE);
double msTime = timer.t;
vector<vector<float> > detections;
for (int k=0; k<100; k++)
{
if(output[7*k+1] == -1)
break;
float classIndex = output[7*k+1];
float confidence = output[7*k+2];
float xmin = output[7*k + 3];
float ymin = output[7*k + 4];
float xmax = output[7*k + 5];
float ymax = output[7*k + 6];
std::cout << classIndex << " , " << confidence << " , " << xmin << " , " << ymin<< " , " << xmax<< " , " << ymax << std::endl;
int x1 = static_cast<int>(xmin * srcImg.cols);
int y1 = static_cast<int>(ymin * srcImg.rows);
int x2 = static_cast<int>(xmax * srcImg.cols);
int y2 = static_cast<int>(ymax * srcImg.rows);
cv::rectangle(srcImg,cv::Rect2f(cv::Point(x1,y1),cv::Point(x2,y2)),cv::Scalar(255,0,255),1);
}
timer.toc();
cv::imshow("mobileNet",srcImg);
cv::waitKey(0);
free(imgData);
cudaFree(imgCUDA);
cudaFreeHost(imgCPU);
cudaFree(output);
tensorNet.destroy();
return 0;
}