我正在尝试在python2中实现cv::cuda::warpPerspective，这里有一篇关于如何做到这一点的非常甜蜜的文章：link。我按照该帖子中所述的说明进行操作，但是，我遇到了Segmentation fault (core dumped)错误。
我能够在文件第 11 行GpuWrapper.pyx分配错误：
pyopencv_to(<PyObject*> _src, src_mat)
似乎它无法将 numpy 数组转换为 opencv Mat。
我不确定哪里出了问题以及如何解决它。
出现Segmentation fault (core dumped)错误的 python 脚本如下所示：
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
另一件事要提一下，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
根据这里的说法，这似乎很好。
环境： 库达 8.0 OpenCV 3.3，无需额外模块 乌班图16.04 蟒蛇 2.7 赛通 0.26
我已经在C++中尝试过cv::cuda::warpPerspective，并且在GPU支持下工作正常。我也尝试了cv2.warpPerspectivepython，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，我将在查找opencv时遇到问题：
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。
更新：我能够使用 pdb 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 }
似乎它不能传递给PyArray_Check，但我不知道发生了什么以及如何解决它。
终于找到了答案。 为了使用numpy，我们必须调用numpy.import_array()。有关详细信息，请参阅此处。
找到在哪里调用它可能很棘手，就我而言，我只需调用我的 .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转。