尝试编写 C++ 包装函数时，无法在 Cython 中将 Numpy 数组转换为 OpenCV Mat

我正在尝试实施cv::cuda::warpPerspective在 python2 中，有一篇关于如何做到这一点的非常精彩的文章：link https://stackoverflow.com/a/42401559/7555390。我按照该帖子中描述的说明进行操作，但是，我得到了Segmentation fault (core dumped) error.

我能够将错误分配给GpuWrapper.pyx文件第 11 行：

pyopencv_to(<PyObject*> _src, src_mat)

似乎无法将 numpy 数组转换为 opencv Mat。

我不确定哪里出了问题以及如何修复。

得到的python脚本Segmentation fault (core dumped)错误如下：

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

另外值得一提的是，在编译 cython 代码时，它有一些警告：

[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

根据这似乎没问题here https://stackoverflow.com/questions/25789055/cython-numpy-warning-about-npy-no-deprecated-api-when-using-memoryview.

环境： CUDA 8.0 OpenCV 3.3 无需额外模块 乌班图16.04 Python 2.7 赛通0.26

我努力了cv::cuda::warpPerspective在 C++ 中，它在 GPU 支持下工作得很好。我也尝试过cv2.warpPerspective在 python 中，没有任何问题（所以我的 python opencv 被正确编译）。

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否则我在寻找 opencv 时会遇到问题：

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，但我也无法解决它。

UPDATE:我能够使用 pdb 来分配错误：

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

似乎它无法将 o 传递给 PyArray_Check，但我不知道发生了什么以及如何修复它。

最后我找到了答案。 为了使用 numpy，我们必须调用numpy.import_array()。详细信息请参见Here https://docs.scipy.org/doc/numpy-1.10.0/user/c-info.how-to-extend.html#required-subroutine.

找到在哪里调用它可能很棘手，对于我的情况，我只是调用我的 .pyx 脚本：

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作。