OpenCV 2.4.10 + Maxwell gpu cmake = <invalid device function> error

asked 2015-08-06 16:55:23 -0600

I installed a NVidia Geforce GT750Ti with Auto and PTX=5.0, ArchBin = 2.0 - 5.0. My CUDA version is 7. I tried out some functions. For some reason, my gpu::transpose and gpu::flip functions runs fine, but my gpu::cvtColor and gpu::resize gives an error.

while (1)
{
    frameNo++;
    int _startWhileLoop = (int)getTickCount();
    capL.read(leftFrame);
    cvtColor(leftFrame, leftFrame, CV_BGR2GRAY);
    imageLSrc.upload(leftFrame);
    cout << "breakpoint 1" << endl;
    cv::gpu::resize(imageLSrc, imageLDst, cv::Size(frameWidth, frameHeight)); // Breaks here
    cout << "breakpoint 2" << endl;
    gpu::transpose(imageLDst, imageLSrc);
    gpu::flip(imageLSrc, imageLDst, 1);
    imageLDst.download(leftFrame);
    cv::imshow("und", leftFrame);

    if (waitKey(30) == 27)
        break;
}

error: OpenCV error: Gpu API call <invalid device="" function=""> in unknown function: ....\src\cuda\resize.cu, line 233

When I try gpu::cvtColor it gives similar error: OpenCV error: Gpu API call <invalid device="" function=""> in unknown function: ....\opencv2\gpu\device\detail\transform_detail.hpp, line 320

Spent alot of time remaking opencv with CUDA. If anyone has made it work with GT750ti, please let me know cmake settings

Thanks

edit retag flag offensive close merge delete