I have a project at hand which I want to use one of the opencv modules (specifically dnn).
Instead of building the dnn module I want to use the source code of this modules in my project. by doing so, I can change the source code live and see the results at the same time.
I have a very simple scenario with one only source file:
main.cpp
#include "iostream"
#include <opencv2/dnn.hpp>
int main(int argc, char *argv[])
{
std::string ConfigFile = "tsproto.pbtxt";
std::string ModelFile = "tsmodel.pb";
cv::dnn::Net net = cv::dnn::readNetFromTensorflow(ModelFile,ConfigFile);
return 0;
}
now this function "cv::dnn::readNetFromTensorflow" is in dnn module. I tried many different methods to embedded the dnn source codes inside my project but all of them failed !
for example, the first time I tried to include every cpp and hpp file in the module/dnn/ folder of opencv in my project but I ended up in errors like
/home/user/projects/Tensor/tf_importer.cpp:28: error: 'CV__DNN_EXPERIMENTAL_NS_BEGIN' does not name a type
#include "../precomp.hpp" no such file or directory
HAVE_PROTOBUF is undefined
and ....
I tried to solve these errors but some more errors just happened, more undefined MACROs and more undefined hpp files !
#include "../layers_common.simd.hpp" no such file or directory
and many many more errors !
It seems that I'm stuck in a while(true) loop of errors !!! Is it really that hard to use opencv modules source code ?
P.S.
For those who are asking about why I want to use opencv source code instead of using the shared libraries I have to say that I want to import a customized tensorflow model which opencv read function doesn't support and I want to know where exactly it crashesh so I can fix it.
By the way, I am only using c++11 functions and gcc as compiler in Ubuntu 16.04
Related
I am trying to run YOLOv3 on Visual Studio 2019 using CUDA 10.2 with cuDNN v7.6.5 on Windows 10 using NVidia GeForce 930M. Here is part of the code I used.
#include <fstream>
#include <sstream>
#include <iostream>
#include <opencv2/dnn.hpp>
#include <opencv2/imgproc.hpp>
#include <opencv2/highgui.hpp>
using namespace cv;
using namespace dnn;
using namespace std;
int main()
{
// Load names of classes
string classesFile = "coco.names";
ifstream ifs(classesFile.c_str());
string line;
while (getline(ifs, line)) classes.push_back(line);
// Give the configuration and weight files for the model
String modelConfiguration = "yolovs.cfg";
String modelWeights = "yolov3.weights";
// Load the network
Net net = readNetFromDarknet(modelConfiguration, modelWeights);
net.setPreferableBackend(DNN_BACKEND_CUDA);
net.setPreferableTarget(DNN_TARGET_CUDA);
// Open the video file
inputFile = "vid.mp4";
cap.open(inputFile);
// Get frame from the video
cap >> frame;
// Create a 4D blob from a frame
blobFromImage(frame, blob, 1 / 255.0, Size(inpWidth, inpHeight), Scalar(0, 0, 0), true, false);
// Sets the input to the network
net.setInput(blob);
// Runs the forward pass to get output of the output layers
vector<Mat> outs;
net.forward(outs, getOutputsNames(net));
}
Although I add $(CUDNN)\include;$(cudnn)\include; to Additional Include Directories in both C/C++ and Linker, added CUDNN_HALF;CUDNN; to C/C++>Preprocessor Definitions, and added cudnn.lib; to Linker>Input, I still get this warning:
DNN module was not built with CUDA backend; switching to CPU
and it runs on CPU instead of GPU, can anyone help me with this problem?
I solved it by using CMake, but I had first to add this opencv_contrib then rebuilding it using Visual Studio. Make sure that these WITH_CUDA, WITH_CUBLAS, WITH_CUDNN, OPENCV_DNN_CUDA, BUILD_opencv_world are checked in CMake.
I had a similar issue happen to me about a week ago, but I was using Python and Tensorflow. Although the languages were different compared to C++, I did get the same error. To fix this, I uninstalled CUDA 10.2 and downgraded to CUDA 10.1. From what I have found, there might be a dependency issue with CUDA, or in your case, OpenCV hasn't created support yet for the latest version of CUDA.
EDIT
After some further research it seems to be an issue with Opencv rather than CUDA. Referencing this github thread, if you installed Opencv with cmake, remove the arch bin version below 7 on the config file, then rebuild/reinstall Opencv. However, if that doesn't work, another option would be to remove CUDA arch bin version < 5.3 and rebuild.
i have installed opencv 3.1.0 and i want to compile following program:
#include <stdio.h>
#include "opencv/cv.h"
#include "opencv/highgui.h"
int main(void){
return 0;
}
but it returns error:undefined reference to cvRound and so on in header file types_c.h.
I know it must be a linking problem but the only libraries i can link are:
opencv_world310.lib, opencv_world310d.lib
I have already tried to link these but that doesn't solve the problem.
I m using a GNU GCC Compiler and dynamic libraries a link with mingw32-g++.exe.
According to some research in the internet i need to link libraries like:
`opencv_calib3d249d.lib`
opencv_contrib249d.lib
opencv_core249d.lib
opencv_features2d249d.lib
opencv_flann249d.lib
opencv_gpu249d.lib
opencv_highgui249d.lib
opencv_imgproc249d.lib
opencv_legacy249d.lib
opencv_ml249d.lib
opencv_nonfree249d.lib
opencv_objdetect249d.lib
opencv_ocl249d.lib
opencv_photo249d.lib
opencv_stitching249d.lib
opencv_superres249d.lib
opencv_ts249d.lib
opencv_video249d.lib
opencv_videostab249d.lib`
but those are not in my lib directory!
I'm trying to work with Opencv CUDA module, specially refer to cv::cuda::log function.
First, I'll give to details Opencv compilation.
I compiled Opencv with WITH_CUDA flag on, took the libs and dlls from the compilation, however I copied the headers files from the downloaded opencv folder, since the compilation folder does't include headers by default.
I wonder, whether is this the right thing to do ?
Second, I tried to use the cv::cuda:: function.
I include the cuda.hpp header
#include "opencv2/core/cuda.hpp"
cv::cuda::GpuMat source, dest;
GpuMat compiles great for me, However I don't know which file should I include in order to work with the log function. when I write the following line
cv::cuda::log(source, dest);
I kept on getting the error message:
error: C2039: log in not a member of cv::cuda
Windows 7, Visual studio 2013, Opencv 3.0.0, platform: 64 bit, CUDA toolkit 6.5
Third, I'd like to know about Opencv CUDA implementation, does it utilize npp functionality? Opencv vs npp, which one is better to use ?
I could easly write my code using npp, however I'd like to know the opencv CUDA module.
Thanks
After couple days of searching, I'd like to share my knowledge
First thing I was doing wrong, was taking the headers from Opencv compilation, the right thing to do is taking the header from all the Opencv modules (each module individually).
Second, After Opencv compilation with CUDA flag everything worked great.
Third, Several opencv CUDA functions does utilize NPP
Forth, Use github
This code should work for OpenCV 3.1:
#include <opencv2/opencv.hpp>
#include <opencv2/cudaarithm.hpp>
int main()
{
cv::Mat img = cv::imread("img.jpg", cv::IMREAD_GRAYSCALE);
cv::Mat img_32f;
img.convertTo(img_32f, CV_32F);
//To avoid log(0) that is undefined
img_32f += 1.0f;
cv::cuda::GpuMat gpuImg, gpuImgLog;
gpuImg.upload(img_32f);
cv::cuda::log(gpuImg, gpuImgLog);
cv::Mat imgLog, imgLog_32f;
gpuImgLog.download(imgLog_32f);
double min, max;
cv::minMaxLoc(imgLog_32f, &min, &max);
imgLog_32f.convertTo(imgLog, CV_8U, 255.0/(max-min), -255.0*min/(max-min));
cv::imshow("img", img);
cv::imshow("imgLog", imgLog);
cv::waitKey(0);
return 0;
}
I have a problem using VideoCapture class with OpenCV 2.4.2 under windows XP 32bits.
It doesn't open any file or camera and fixing it's being a pain.
Im using visual studio 2010 but i have also tried the code in QTcreator with the same result.
The testing code is the following:
#include "opencv/cv.h"
#include "opencv/highgui.h"
#include <iostream>
#include <string>
#include <iomanip>
#include <sstream>
using namespace cv;
using namespace std;
int main()
{
const char* videoPath = "C:/video/";
string videoName = string(videoPath) + "avi.avi";
VideoCapture cap(videoName);
if(!cap.isOpened())
{
std::cout<<"Fail"<<std::endl;
return -3;
}
return 0;
}
The output is always '-3'.
Qt Creator shows a
warning: Error opening file (../../modules/highgui/src/cap_ffmpeg_impl.hpp:361)
I debugged it and the problem appears in the first line of:
CvCapture* cvCreateFileCapture_FFMPEG_proxy(const char * filename)
{
CvCapture_FFMPEG_proxy* result = new CvCapture_FFMPEG_proxy;
if( result->open( filename ))
return result;
delete result;
#if defined WIN32 || defined _WIN32
return cvCreateFileCapture_VFW(filename);
#else
return 0;
#endif
}
in the cap_ffmpeg.cpp internal file.
I have tested the same code in a mac under snow leopard and it works. No surprises here since it must be a library issue.
I have opened the avi file with the same path route using the c-function cvCapture easy and fast.
I got all the dlls of 'C:\opencv\opencv\build\x86\vc10\bin'
included in mi debug file. I got the tbb.dll and all the 'C:\opencv\opencv\3rdparty\ffmpeg' content included too.
This is drving me crazy so any help would be appreciated.
Thanks in advance.
In my case, the same problem was resolved after deleting all opencv_***.dll files in C:\Windows\System32. So, I use the dll files just through the path like "%PATH%;C: \Program Files \OpenCV2.4.2\build\x86\vc10/bin". Please try it.
I also faced with this problem and solved it by correct the path of the function:
VideoCapture cap(videoName);
If the AVI file of videoName does't exist, it will be an error:
(../../modules/highgui/src/cap_ffmpeg_impl.hpp:XXX)
where XXX represents the line number.
I had the same issue with the open method whilst running under Windows 8 (64bit), opencv 2.4.10. IDE is running in x86.
I found that running the application in release configuration solved the problem.
Stumbled across the answer because I had the same issue with imread. Issue is presented in the this thread.
imread not working in Opencv
See the fix I found below, for mp4 files.
I faced the same issue on Windows 7, using OpenCV 2.4.9. I am using the java wrapper for opencv.
Matthias Krings has done a lot of research for this. See this. Apparently this is an issue based on the video file type. With .avi files, it seems to work for a lot of people. Unfortunately his solution of setting OPENCV_DIR did not work for me. But his comments in the bug listing gave me a hint to fix the issue.
You have to do two things:
Set java.library.path to include the directory {opencv\install\dir}opencv-2.4.9\build\x86\vc10\bin. You can set the variable using the -D option on the java command line: java -Djava.library.path=PATH_TO_YOUR_DLL .... Also fetch this variable from your environment, using System.getProperty(...), and print it before calling loadLibrary(), to verify that the path setting is working.
And in your java class, load the ffmpeg dll using System.loadLibrary("opencv_ffmpeg249");. The loadLibrary() function should be invoked from within a static block in java.
There is a file named opencv_ffmpeg249.dll in the java.library.path that we set.
This works on ubuntu also, for .so files.
I too faced the same issue and resolved after pointing to the correct location of the input video.
Here's my code (the first DisplayImage.cpp code in the OpenCV documentation)
/*
* DisplayImage.cpp
*
* Created on: Dec 25, 2011
* Author: Arcturus */
#include <iostream>
#include <opencv2\opencv.hpp>
using namespace cv;
using namespace std;
int main(int argc, char** argv){
Mat image;
image = imread(argv[1], 1);
if(argc!=2 || !image.data){
cout<<"no image data";
return -1;
}
namedWindow("Display Image", CV_WINDOW_AUTOSIZE);
imshow("Display Image", image);
waitKey(10000);
return 0;
}
Build complete, executable generated, binaries generated.
I have my image - blackbuck.bmp- in the DisplayImage Debug folder. To run the code, I go to Run> Run Configurations. Select the DisplayImage Debug exe file, key in blackbuck.bmp (also tried it with absolute path) and run it.
On the top of the console, I get the message : DisplayImage Debug. And it displays no image at all. What could be wrong here?
I am running it on Eclipse, using CDT.
Thank you for your time!
EDIT: Problem solved!!! I had to copy all the dll files from the library folder to the folder in which my executable file was being generated. I still do not understand why, though. After all, the linker was already linking the library folder containing all the dlls. If someone could explain this, it would be of great help for future debugging. Thank you karl and mevotron for your time :)
EDIT 2: From the msdn website:
"A potential disadvantage to using DLLs is that the application is not self-contained; it depends on the existence of a separate DLL module. The system terminates processes using load-time dynamic linking if they require a DLL that is not found at process startup and gives an error message to the user. The system does not terminate a process using run-time dynamic linking in this situation, but functions exported by the missing DLL are not available to the program."
I think this answers my question. Perhaps this means eclipse uses load-time dynamic linking.
How did you compile OpenCV with MinGW (i.e., what were your BUILD_TYPE and SSE* options set to during the CMake configuration)? The reason I ask, is that there is a known bug with SSE optimizations that will cause highgui operations to crash when using MinGW built versions. See my other SO answer here.