尝试编写C ++包装函数时无法将Numpy数组转换为Cython中的OpenCV Mat [英] Not able to convert Numpy array to OpenCV Mat in Cython when trying to write c++ wrapper function

问题描述

我正在尝试在python2中实现cv::cuda::warpPerspective，在这里有一个关于如何执行此操作的非常不错的文章:链接.我按照该帖子中的说明进行了操作，但是，出现了Segmentation fault (core dumped)错误.

I am trying to implement cv::cuda::warpPerspective in python2, there is a very sweet post about how to do that here: link. I followed the instruction as described in that post, however, I got Segmentation fault (core dumped) error.

我能够在GpuWrapper.pyx文件第11行中分配错误:

I was able to allocate the error in GpuWrapper.pyx file line 11:

pyopencv_to(<PyObject*> _src, src_mat)

似乎无法将numpy数组转换为opencv Mat.

It seems that it fails to convert numpy array to opencv Mat.

我不确定哪里出了问题以及如何解决.

I am not sure where is wrong and how to fix it.

出现Segmentation fault (core dumped)错误的python脚本如下:

The python script that got Segmentation fault (core dumped) error is in below:

import cv2
import numpy as np
import csv
import timeit
import GpuWrapper

while (1):
    start = timeit.default_timer()
    eo_img = cv2.imread('./sample/eo.png', 1)
    nir_img = cv2.imread('./sample/nir.png', 0)
    with open('./sample/reg.csv', 'rb') as f:
        line = csv.reader(f)
        reg_line = list(line)
    reg = np.array(reg_line[0], dtype=np.float32)
    new_reg = reg.reshape((3,3))
    print nir_img.shape
    dist = GpuWrapper.cudaWarpPerspectiveWrapper(nir_img, new_reg, (2448,2048))
    cv2.imwrite('./sample/result.png', dist)
    end = timeit.default_timer()
    print end-start

要提及的另一件事是，在编译cython代码时，它会发出一些警告:

Another thing to mention, when compiling cython code, it has some warnings:

[1/1] Cythonizing GpuWrapper.pyx
running build_ext
building 'GpuWrapper' extension
x86_64-linux-gnu-gcc -pthread -DNDEBUG -g -fwrapv -O2 -Wall -Wstrict-prototypes -fno-strict-aliasing -Wdate-time -D_FORTIFY_SOURCE=2 -g -fstack-protector-strong -Wformat -Werror=format-security -fPIC -I/home/xinyao/Tensorflow/local/lib/python2.7/site-packages/numpy/core/include -I-I/usr/local/include/opencv -I-I/usr/local/include -I/usr/include/python2.7 -c GpuWrapper.cpp -o build/temp.linux-x86_64-2.7/GpuWrapper.o
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++
In file included from /home/xinyao/Tensorflow/local/lib/python2.7/site-packages/numpy/core/include/numpy/ndarraytypes.h:1809:0,
                 from /home/xinyao/Tensorflow/local/lib/python2.7/site-packages/numpy/core/include/numpy/ndarrayobject.h:18,
                 from pyopencv_converter.cpp:2,
                 from GpuWrapper.cpp:544:
/home/xinyao/Tensorflow/local/lib/python2.7/site-packages/numpy/core/include/numpy/npy_1_7_deprecated_api.h:15:2: warning: #warning "Using deprecated NumPy API, disable it by " "#defining NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION" [-Wcpp]
 #warning "Using deprecated NumPy API, disable it by " \
  ^
c++ -pthread -shared -Wl,-O1 -Wl,-Bsymbolic-functions -Wl,-Bsymbolic-functions -Wl,-z,relro -fno-strict-aliasing -DNDEBUG -g -fwrapv -O2 -Wall -Wstrict-prototypes -Wdate-time -D_FORTIFY_SOURCE=2 -g -fstack-protector-strong -Wformat -Werror=format-security -Wl,-Bsymbolic-functions -Wl,-z,relro -Wdate-time -D_FORTIFY_SOURCE=2 -g -fstack-protector-strong -Wformat -Werror=format-security build/temp.linux-x86_64-2.7/GpuWrapper.o -o /home/xinyao/projects/image_reg/GpuWrapper.so -L/usr/local/lib -lopencv_cudabgsegm -lopencv_cudaobjdetect -lopencv_cudastereo -lopencv_dnn -lopencv_ml -lopencv_shape -lopencv_stitching -lopencv_cudafeatures2d -lopencv_superres -lopencv_cudacodec -lopencv_videostab -lopencv_cudaoptflow -lopencv_cudalegacy -lopencv_calib3d -lopencv_features2d -lopencv_highgui -lopencv_videoio -lopencv_photo -lopencv_imgcodecs -lopencv_cudawarping -lopencv_cudaimgproc -lopencv_cudafilters -lopencv_video -lopencv_objdetect -lopencv_imgproc -lopencv_flann -lopencv_cudaarithm -lopencv_viz -lopencv_core -lopencv_cudev

根据这里.

环境: CUDA 8.0 无需额外模块的OpenCV 3.3 Ubuntu 16.04 Python 2.7 Cython 0.26

Environment: CUDA 8.0 OpenCV 3.3 without extra modules Ubuntu 16.04 Python 2.7 Cython 0.26

我在C ++中尝试了cv::cuda::warpPerspective，它在GPU支持下工作正常.我也在python中试过cv2.warpPerspective，没有任何问题(所以我的python的opencv正确编译了.)

I have tried cv::cuda::warpPerspective in C++ and it works fine with GPU support. I have also tried cv2.warpPerspective in python and without any issue(so my opencv for python was compiled correctly).

另一件事是我不能直接用g ++编译opencv C ++代码，我需要添加额外的标志来做到这一点，就像这样: g++ -o main main.cc `pkg-config opencv --cflags --libs 否则，我在找到opencv时会遇到问题:

Another thing is I cannot compile opencv C++ code directly with g++, I need to add extra flags to do it, like in this way: g++ -o main main.cc `pkg-config opencv --cflags --libs Otherwise I will have issues on finding opencv:

main.cc:(.text+0x45): undefined reference to `cv::cuda::GpuMat::defaultAllocator()'
main.cc:(.text+0x9b): undefined reference to `cv::imread(cv::String const&, int)'
main.cc:(.text+0x104): undefined reference to `cv::imread(cv::String const&, int)'
main.cc:(.text+0x140): undefined reference to `cv::cuda::GpuMat::defaultAllocator()'
main.cc:(.text+0x1a5): undefined reference to `cv::Mat::eye(int, int, int)'
main.cc:(.text+0x31f): undefined reference to `cv::cuda::Stream::Null()'
main.cc:(.text+0x3d6): undefined reference to `cv::cuda::warpPerspective(cv::_InputArray const&, cv::_OutputArray const&, cv::_InputArray const&, cv::Size_<int>, int, int, cv::Scalar_<double>, cv::cuda::Stream&)'
main.cc:(.text+0x487): undefined reference to `cv::imwrite(cv::String const&, cv::_InputArray const&, std::vector<int, std::allocator<int> > const&)'

不确定这是否是一个因素，但我也无法修复.

Not sure if that is a factor, but I am not able to fix it as well.

更新: 我可以使用pdb分配错误:

UPDATE: I was able to allocate the bug by using pdb:

>$gdb --args python opencv_reg_cuda.py
>$run

    Starting program: /usr/bin/python opencv_reg_cuda.py
    [Thread debugging using libthread_db enabled]
    Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
    [New Thread 0x7fffc6b83700 (LWP 10933)]
    [New Thread 0x7fffc4382700 (LWP 10934)]
    [New Thread 0x7fffc1b81700 (LWP 10935)]
    [New Thread 0x7fffc1380700 (LWP 10936)]
    [New Thread 0x7fffbcb7f700 (LWP 10937)]
    [New Thread 0x7fffba37e700 (LWP 10938)]
    [New Thread 0x7fffb7b7d700 (LWP 10939)]
    (2048, 2448)

    Thread 1 "python" received signal SIGSEGV, Segmentation fault.
    pyopencv_to (o=o@entry=0x7fffb37e3b70, m=..., info=...) at pyopencv_converter.cpp:210
    210     if( !PyArray_Check(o) )

>$bt

    #0  pyopencv_to (o=o@entry=0x7fffb37e3b70, m=..., info=...) at pyopencv_converter.cpp:210
    #1  0x00007fffb33c56f5 in pyopencv_to<cv::Mat> (name=0x7fffb33c98d8 "<unknown>", m=..., o=0x7fffb37e3b70) at pyopencv_converter.cpp:349
    #2  __pyx_pf_10GpuWrapper_cudaWarpPerspectiveWrapper (__pyx_v(short, long double, char)=__pyx_v(short, long double, char)@entry=0x7fffb37e3b70, __pyx_v__M=__pyx_v__M@entry=0x7fffb37e3c60, 
        __pyx_v__size_tuple=__pyx_v__size_tuple@entry=0x7ffff7e80b90, __pyx_v__flags=<optimized out>, __pyx_self=<optimized out>) at GpuWrapper.cpp:1796
    #3  0x00007fffb33c76a2 in __pyx_pw_10GpuWrapper_1cudaWarpPerspectiveWrapper (__pyx_self=<optimized out>, __pyx_args=<optimized out>, __pyx_kwds=<optimized out>) at GpuWrapper.cpp:1737
    #4  0x00000000004c468a in PyEval_EvalFrameEx ()
    #5  0x00000000004c2765 in PyEval_EvalCodeEx ()
    #6  0x00000000004c2509 in PyEval_EvalCode ()
    #7  0x00000000004f1def in ?? ()
    #8  0x00000000004ec652 in PyRun_FileExFlags ()
    #9  0x00000000004eae31 in PyRun_SimpleFileExFlags ()
    #10 0x000000000049e14a in Py_Main ()
    #11 0x00007ffff7810830 in __libc_start_main (main=0x49dab0 <main>, argc=2, argv=0x7fffffffde38, init=<optimized out>, fini=<optimized out>, rtld_fini=<optimized out>, stack_end=0x7fffffffde28)
        at ../csu/libc-start.c:291
    #12 0x000000000049d9d9 in _start ()

>list

    205             }
    206         }
    207         return true;
    208     }
    209 
    210     if( !PyArray_Check(o) )
    211     {
    212         failmsg("%s is not a numpy array, neither a scalar", info.name);
    213         return false;
    214     }

似乎无法将o传递给PyArray_Check，但我不知道发生了什么以及如何解决.

It seems like it cannot pass o to PyArray_Check, but I have no idea what was going on and how to fix it.

推荐答案

最后我找到了答案. 为了使用numpy，我们必须调用numpy.import_array().有关详细信息，请参见

Finally I found the answer. In order to use numpy, we have to call numpy.import_array(). For detail, see Here.

找到调用它的位置可能很棘手，就我而言，我只是在.pyx脚本中调用:

It can be tricky to find where to call it, for my case, I just call in my .pyx script:

import numpy as np  # Import Python functions, attributes, submodules of numpy
cimport numpy as np  # Import numpy C/C++ API

np.import_array()

def cudaWarpPerspectiveWrapper(np.ndarray[np.uint8_t, ndim=2] _src,
                               np.ndarray[np.float32_t, ndim=2] _M,
                               _size_tuple,
                               int _flags=INTER_NEAREST):
    # Create GPU/device InputArray for src

    cdef Mat src_mat
    cdef GpuMat src_gpu
    pyopencv_to(<PyObject*> _src, src_mat)
    src_gpu.upload(src_mat)

    # Create CPU/host InputArray for M
    cdef Mat M_mat = Mat()
    pyopencv_to(<PyObject*> _M, M_mat)

    # Create Size object from size tuple
    # Note that size/shape in Python is handled in row-major-order -- therefore, width is [1] and height is [0]
    cdef Size size = Size(<int> _size_tuple[0], <int> _size_tuple[1])

    # Create empty GPU/device OutputArray for dst
    cdef GpuMat dst_gpu = GpuMat()
    warpPerspective(src_gpu, dst_gpu, M_mat, size, 2)

    # Get result of dst
    cdef Mat dst_host
    dst_gpu.download(dst_host)
    cdef np.ndarray out = <np.ndarray> pyopencv_from(dst_host)
    return out

然后一切都像魔术一样运转.

Then everything is working like a magic.

这篇关于尝试编写C ++包装函数时无法将Numpy数组转换为Cython中的OpenCV Mat的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！