opencv 2410 cascade_gpu corrupt

asked 2018-07-17 23:43:31 -0600

updated 2018-07-17 23:45:47 -0600

berak gravatar image

image description

hi .guys when i use this model to detect obj, the first time i run it is ok then if i release this cascade_gpu model and renew allocate it, it would break like the picture show.

i trace the source code and found it break here :

template <bool tbDoSqr, class T_in, class T_out>
NCVStatus scanRowsWrapperDevice(T_in *d_src, Ncv32u srcStride,
                                T_out *d_dst, Ncv32u dstStride, NcvSize32u roi)
{
    cudaChannelFormatDesc cfdTex;
........
    scanRows
        <T_in, T_out, tbDoSqr>
        <<<roi.height, NUM_SCAN_THREADS, 0, nppStGetActiveCUDAstream()>>>
        (d_src, (Ncv32u)alignmentOffset, roi.width, srcStride, d_dst, dstStride);

    ncvAssertCUDALastErrorReturn(NPPST_CUDA_KERNEL_EXECUTION_ERROR);

    return NPPST_SUCCESS;
}

and i compare the codes between 2410 and 2413 and found difference in scanRows founction.but recompile the opencv_gpu2410.lib still can not fix this problem.

could anyone please tell me the way out?

edit retag flag offensive close merge delete

Comments

1

the 2.4 branch is no more developed since 5 years (bugfixes only)

please consider updating to latest 3.4.2

berak gravatar imageberak ( 2018-07-17 23:49:01 -0600 )edit

On top of that, the GPU cascades have been broken for at least 4 years :D one should rewrite the interface, but with deep learning taking over, no-one has bothered.

StevenPuttemans gravatar imageStevenPuttemans ( 2018-07-24 04:53:02 -0600 )edit