GPU Coder Image output?

2 次查看(过去 30 天)
Joseph Rivera
Joseph Rivera 2020-3-30
I'm using gpu coder to generate the function below, modified from the resnet50_demo, to run a yolo network to a jetson Nano.
function out = resnet50_wrapper(im)
persistent yolov2Obj;
opencv_link_flags = '`pkg-config --cflags --libs opencv`';
coder.updateBuildInfo('addLinkFlags',opencv_link_flags);
if isempty(yolov2Obj)
yolov2Obj = coder.loadDeepLearningNetwork('C:\Users\josep\Desktop\Fall_2019\Senior_Design\Matlab_Tools\yolov2_5eps_128.mat');
end
% pass in input
% [bboxes,scores,labels] = yolov2Obj.detect(im,'Threshold',0.5);
[bboxes,~,labels] = yolov2Obj.detect(im,'Threshold',0.5);
% Annotate detections in the image.
out = insertObjectAnnotation(im,'rectangle',bboxes,labels);
end
I believe the code generation for the function works fine, but what's the best way to display the output image? I am editting the custom main function from the resnet demo, the function returns the 'out' image as a real32_T, how do I convert it to a mat to display with openCV? The line I'm thinking of adding is just a type conversion underneath "convert out -> oFrame".
/* The MathWorks Inc. 2019*/
/* ResNet50 demo main.cu file with OpenCV interfaces to read and display data. */
#include "resnet50_wrapper.h"
#include "main.h"
#include "resnet50_wrapper_terminate.h"
#include "resnet50_wrapper_initialize.h"
#include "opencv2/opencv.hpp"
#include <stdio.h>
#include <stdlib.h>
#define IMG_WIDTH 224
#define IMG_HEIGHT 224
#define IMG_CH 3
#define VID_DEV_ID 1
using namespace cv;
using namespace std;
static void main_resnet50_wrapper();
/*
* Convert BGR data to RGB data, without this conversion the predictions
* will be bad
*/
static void argInit_224x224x3_real32_T(real32_T *input, Mat & im)
{
for(int j=0;j<224*224;j++)
{
//BGR to RGB
input[2*224*224+j]=(float)(im.data[j*3+0]);
input[1*224*224+j]=(float)(im.data[j*3+1]);
input[0*224*224+j]=(float)(im.data[j*3+2]);
}
}
static void main_resnet50_wrapper(void)
{
static real32_T b[150528];
static real32_T out[150528];
Mat oFrame, cFrame;
/* Create a Video capture object */
VideoCapture cap(VID_DEV_ID);
if(!cap.isOpened())
{
cout << "can't open camera" << endl;
exit(0);
}
namedWindow("resnet Demo",CV_WINDOW_NORMAL);
resizeWindow("resnet Demo", 1000,1000);
float fps=0;
cudaEvent_t start, stop;
cudaEventCreate(&start);
cudaEventCreate(&stop);
while(1)
{
cap >> oFrame;
resize(oFrame,cFrame,Size(IMG_WIDTH,IMG_HEIGHT));
/* convert from BGR to RGB*/
argInit_224x224x3_real32_T(b,cFrame);
cudaEventRecord(start);
/* call the resnet predict function*/
// resnet50_wrapper(b, out);
resnet50_wrapper(b,out);
cudaEventRecord(stop);
cudaEventSynchronize(stop);
float milliseconds = -1.0;
cudaEventElapsedTime(&milliseconds, start, stop);
fps = fps*.9+1000.0/milliseconds*.1;
// convert out -> oFrame
// show display
imshow("Yolo Demo", oFrame);
if(waitKey(1)%256 == 27 ) break; // stop when ESC key is pressed
}
}
int32_T main(int32_T argc, const char * const argv[])
{
(void)argc;
(void)argv;
/* Call the application intialize function */
resnet50_wrapper_initialize();
/* Call the resnet predict function */
main_resnet50_wrapper();
/* Call the application terminate function */
resnet50_wrapper_terminate();
return 0;
}

回答(1 个)

Dinesh Iyer
Dinesh Iyer 2020-3-31
Joseph,
A few things that might help. The cv::Mat's can store data in different color formats. The most popular one that I have seen is BGRA. You can confirm this by inspecting oFrame.channels() and oFrame.depth(). So, your conversion routine argInit_224x224x3_real32_T probably needs to be modified to perform the copy by inspecting the im.channels() and possibly im.depth(). Converting from RGB to cv::Mat is the exact inverse operation. But try to convert it to BGRA.
Hope this helps.
Dinesh

类别

Help CenterFile Exchange 中查找有关 Computer Vision Toolbox Supported Hardware 的更多信息

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by