将ndarray转换为cv::Mat的最简单方法是什么?
我正在尝试为一个使用了OpenCV中的cv::Mat
类的C++库创建一个Python/Cython的封装。在官方的Python封装中,所有的函数都使用NumPy的ndarray
,而不是cv::Mat
,这样用起来非常方便。但是在我自己的封装中,我该如何进行这样的转换呢?也就是说,我该如何从np.ndarray
创建 cv::Mat
?
5 个回答
根据tlorieul的回答,这里是我用来构建Python/C++模块的代码:
https://gist.github.com/des0ps/88f1332319867a678a74bdbc0e7401c2
这个代码已经在Python3和OpenCV3上测试过了。
如果这对你有帮助,我写了一个工具,它正好可以做到这一点。这个工具是一个方便的库,它注册了一个boost::python的转换器,可以自动把OpenCV中常用的cv::Mat数据类型和NumPy中常用的np.array()数据类型互相转换。这样,开发者就可以轻松地在OpenCV的C++接口和用NumPy编写的Python接口之间来回切换,而不需要额外编写处理PyObjects(Python对象)的包装代码。
你可以看看这个链接: https://github.com/spillai/numpy-opencv-converter
我想你可以直接使用或者借鉴一下官方Python封装中的转换器的一些逻辑。这个模块的文档不多,但也许封装生成器的输出能帮助你理解怎么使用它。
其实,把任何一个 np.ndarray
转换成对应的 cv::Mat
并没有简单的方法。基本上,你只需要做两件事:
- 创建一个大小和类型对应的空
cv::Mat
。 - 复制数据。
不过,问题的关键在于细节。ndarray
和 Mat
可能会有很不同的数据格式。例如,NumPy 数组中的数据可能是按照 C 语言的顺序或 Fortran 的顺序排列的,数组对象可能拥有自己的数据,或者只是指向另一个数组的视图,通道的顺序也可能不同(在 NumPy 中是 RGB,而在 OpenCV 中是 BGR)等等。
所以,我决定不去解决这个通用的问题,而是写一些简单的代码,满足我的需求,并且任何感兴趣的人都可以轻松修改。
下面的 Cython 代码可以处理 float32
/CV_32FC1
格式的图像,且使用默认的字节顺序:
cdef void array2mat(np.ndarray arr, Mat& mat):
cdef int r = arr.shape[0]
cdef int c = arr.shape[1]
cdef int mat_type = CV_32FC1 # or CV_64FC1, or CV_8UC3, or whatever
mat.create(r, c, mat_type)
cdef unsigned int px_size = 4 # 8 for single-channel double image or
# 1*3 for three-channel uint8 image
memcpy(mat.data, arr.data, r*c*px_size)
在 Cython 中使用这段代码时,还需要声明一些类型和常量,比如这样:
import numpy as np
# Cython makes it simple to import NumPy
cimport numpy as np
# OpenCV's matrix class
cdef extern from "opencv2/opencv.hpp" namespace "cv":
cdef cppclass Mat:
Mat() except +
Mat(int, int, int, void*) except +
void create(int, int, int)
void* data
int type() const
int cols
int rows
int channels()
Mat clone() const
# some OpenCV matrix types
cdef extern from "opencv2/opencv.hpp":
cdef int CV_8UC3
cdef int CV_8UC1
cdef int CV_32FC1
cdef int CV_64FC1
从 cv::Mat
转换到 np.ndarray
的方法也可以用类似的方式实现。
附加信息:还有一篇不错的 博客文章,介绍了 RGB/BGR 图像的相同类型转换。
正如kyamagu所建议的,你可以使用OpenCV的官方Python封装代码,特别是pyopencv_to
和pyopencv_from
这两个函数。
我曾经和你一样,为了处理所有的依赖关系和生成的头文件而苦恼。不过,通过“清理”这个cv2.cpp
文件,可以减少复杂性,正如lightalchemist在这里所做的那样,只保留必要的部分。你需要根据自己的需求和使用的OpenCV版本进行调整,但基本上我用的代码就是这个。
#include <Python.h>
#include "numpy/ndarrayobject.h"
#include "opencv2/core/core.hpp"
static PyObject* opencv_error = 0;
static int failmsg(const char *fmt, ...)
{
char str[1000];
va_list ap;
va_start(ap, fmt);
vsnprintf(str, sizeof(str), fmt, ap);
va_end(ap);
PyErr_SetString(PyExc_TypeError, str);
return 0;
}
class PyAllowThreads
{
public:
PyAllowThreads() : _state(PyEval_SaveThread()) {}
~PyAllowThreads()
{
PyEval_RestoreThread(_state);
}
private:
PyThreadState* _state;
};
class PyEnsureGIL
{
public:
PyEnsureGIL() : _state(PyGILState_Ensure()) {}
~PyEnsureGIL()
{
PyGILState_Release(_state);
}
private:
PyGILState_STATE _state;
};
#define ERRWRAP2(expr) \
try \
{ \
PyAllowThreads allowThreads; \
expr; \
} \
catch (const cv::Exception &e) \
{ \
PyErr_SetString(opencv_error, e.what()); \
return 0; \
}
using namespace cv;
static PyObject* failmsgp(const char *fmt, ...)
{
char str[1000];
va_list ap;
va_start(ap, fmt);
vsnprintf(str, sizeof(str), fmt, ap);
va_end(ap);
PyErr_SetString(PyExc_TypeError, str);
return 0;
}
static size_t REFCOUNT_OFFSET = (size_t)&(((PyObject*)0)->ob_refcnt) +
(0x12345678 != *(const size_t*)"\x78\x56\x34\x12\0\0\0\0\0")*sizeof(int);
static inline PyObject* pyObjectFromRefcount(const int* refcount)
{
return (PyObject*)((size_t)refcount - REFCOUNT_OFFSET);
}
static inline int* refcountFromPyObject(const PyObject* obj)
{
return (int*)((size_t)obj + REFCOUNT_OFFSET);
}
class NumpyAllocator : public MatAllocator
{
public:
NumpyAllocator() {}
~NumpyAllocator() {}
void allocate(int dims, const int* sizes, int type, int*& refcount,
uchar*& datastart, uchar*& data, size_t* step)
{
PyEnsureGIL gil;
int depth = CV_MAT_DEPTH(type);
int cn = CV_MAT_CN(type);
const int f = (int)(sizeof(size_t)/8);
int typenum = depth == CV_8U ? NPY_UBYTE : depth == CV_8S ? NPY_BYTE :
depth == CV_16U ? NPY_USHORT : depth == CV_16S ? NPY_SHORT :
depth == CV_32S ? NPY_INT : depth == CV_32F ? NPY_FLOAT :
depth == CV_64F ? NPY_DOUBLE : f*NPY_ULONGLONG + (f^1)*NPY_UINT;
int i;
npy_intp _sizes[CV_MAX_DIM+1];
for( i = 0; i < dims; i++ )
_sizes[i] = sizes[i];
if( cn > 1 )
{
/*if( _sizes[dims-1] == 1 )
_sizes[dims-1] = cn;
else*/
_sizes[dims++] = cn;
}
PyObject* o = PyArray_SimpleNew(dims, _sizes, typenum);
if(!o)
CV_Error_(CV_StsError, ("The numpy array of typenum=%d, ndims=%d can not be created", typenum, dims));
refcount = refcountFromPyObject(o);
npy_intp* _strides = PyArray_STRIDES(o);
for( i = 0; i < dims - (cn > 1); i++ )
step[i] = (size_t)_strides[i];
datastart = data = (uchar*)PyArray_DATA(o);
}
void deallocate(int* refcount, uchar*, uchar*)
{
PyEnsureGIL gil;
if( !refcount )
return;
PyObject* o = pyObjectFromRefcount(refcount);
Py_INCREF(o);
Py_DECREF(o);
}
};
NumpyAllocator g_numpyAllocator;
enum { ARG_NONE = 0, ARG_MAT = 1, ARG_SCALAR = 2 };
static int pyopencv_to(const PyObject* o, Mat& m, const char* name = "<unknown>", bool allowND=true)
{
if(!o || o == Py_None)
{
if( !m.data )
m.allocator = &g_numpyAllocator;
return true;
}
if( PyInt_Check(o) )
{
double v[] = {PyInt_AsLong((PyObject*)o), 0., 0., 0.};
m = Mat(4, 1, CV_64F, v).clone();
return true;
}
if( PyFloat_Check(o) )
{
double v[] = {PyFloat_AsDouble((PyObject*)o), 0., 0., 0.};
m = Mat(4, 1, CV_64F, v).clone();
return true;
}
if( PyTuple_Check(o) )
{
int i, sz = (int)PyTuple_Size((PyObject*)o);
m = Mat(sz, 1, CV_64F);
for( i = 0; i < sz; i++ )
{
PyObject* oi = PyTuple_GET_ITEM(o, i);
if( PyInt_Check(oi) )
m.at<double>(i) = (double)PyInt_AsLong(oi);
else if( PyFloat_Check(oi) )
m.at<double>(i) = (double)PyFloat_AsDouble(oi);
else
{
failmsg("%s is not a numerical tuple", name);
m.release();
return false;
}
}
return true;
}
if( !PyArray_Check(o) )
{
failmsg("%s is not a numpy array, neither a scalar", name);
return false;
}
bool needcopy = false, needcast = false;
int typenum = PyArray_TYPE(o), new_typenum = typenum;
int type = typenum == NPY_UBYTE ? CV_8U :
typenum == NPY_BYTE ? CV_8S :
typenum == NPY_USHORT ? CV_16U :
typenum == NPY_SHORT ? CV_16S :
typenum == NPY_INT ? CV_32S :
typenum == NPY_INT32 ? CV_32S :
typenum == NPY_FLOAT ? CV_32F :
typenum == NPY_DOUBLE ? CV_64F : -1;
if( type < 0 )
{
if( typenum == NPY_INT64 || typenum == NPY_UINT64 || type == NPY_LONG )
{
needcopy = needcast = true;
new_typenum = NPY_INT;
type = CV_32S;
}
else
{
failmsg("%s data type = %d is not supported", name, typenum);
return false;
}
}
int ndims = PyArray_NDIM(o);
if(ndims >= CV_MAX_DIM)
{
failmsg("%s dimensionality (=%d) is too high", name, ndims);
return false;
}
int size[CV_MAX_DIM+1];
size_t step[CV_MAX_DIM+1], elemsize = CV_ELEM_SIZE1(type);
const npy_intp* _sizes = PyArray_DIMS(o);
const npy_intp* _strides = PyArray_STRIDES(o);
bool ismultichannel = ndims == 3 && _sizes[2] <= CV_CN_MAX;
for( int i = ndims-1; i >= 0 && !needcopy; i-- )
{
// these checks handle cases of
// a) multi-dimensional (ndims > 2) arrays, as well as simpler 1- and 2-dimensional cases
// b) transposed arrays, where _strides[] elements go in non-descending order
// c) flipped arrays, where some of _strides[] elements are negative
if( (i == ndims-1 && (size_t)_strides[i] != elemsize) ||
(i < ndims-1 && _strides[i] < _strides[i+1]) )
needcopy = true;
}
if( ismultichannel && _strides[1] != (npy_intp)elemsize*_sizes[2] )
needcopy = true;
if (needcopy)
{
if( needcast )
o = (PyObject*)PyArray_Cast((PyArrayObject*)o, new_typenum);
else
o = (PyObject*)PyArray_GETCONTIGUOUS((PyArrayObject*)o);
_strides = PyArray_STRIDES(o);
}
for(int i = 0; i < ndims; i++)
{
size[i] = (int)_sizes[i];
step[i] = (size_t)_strides[i];
}
// handle degenerate case
if( ndims == 0) {
size[ndims] = 1;
step[ndims] = elemsize;
ndims++;
}
if( ismultichannel )
{
ndims--;
type |= CV_MAKETYPE(0, size[2]);
}
if( ndims > 2 && !allowND )
{
failmsg("%s has more than 2 dimensions", name);
return false;
}
m = Mat(ndims, size, type, PyArray_DATA(o), step);
if( m.data )
{
m.refcount = refcountFromPyObject(o);
if (!needcopy)
{
m.addref(); // protect the original numpy array from deallocation
// (since Mat destructor will decrement the reference counter)
}
};
m.allocator = &g_numpyAllocator;
return true;
}
static PyObject* pyopencv_from(const Mat& m)
{
if( !m.data )
Py_RETURN_NONE;
Mat temp, *p = (Mat*)&m;
if(!p->refcount || p->allocator != &g_numpyAllocator)
{
temp.allocator = &g_numpyAllocator;
ERRWRAP2(m.copyTo(temp));
p = &temp;
}
p->addref();
return pyObjectFromRefcount(p->refcount);
}
一旦你有了一个清理过的cv2.cpp
文件,这里有一些Cython代码可以处理转换。注意import_array()
函数的定义和调用(这是一个在cv2.cpp
中某个地方包含的NumPy函数),这个调用是必要的,因为它定义了一些pyopencv_to
使用的宏。如果你不调用它,你会遇到段错误,正如lightalchemist指出的那样。
from cpython.ref cimport PyObject
# Declares OpenCV's cv::Mat class
cdef extern from "opencv2/core/core.hpp":
cdef cppclass Mat:
pass
# Declares the official wrapper conversion functions + NumPy's import_array() function
cdef extern from "cv2.cpp":
void import_array()
PyObject* pyopencv_from(const _Mat&)
int pyopencv_to(PyObject*, _Mat&)
# Function to be called at initialization
cdef void init():
import_array()
# Python to C++ conversion
cdef Mat nparrayToMat(object array):
cdef Mat mat
cdef PyObject* pyobject = <PyObject*> array
pyopencv_to(pyobject, mat)
return <Mat> mat
# C++ to Python conversion
cdef object matToNparray(Mat mat):
return <object> pyopencv_from(mat)
注意:我在Fedora 20上使用NumPy 1.8.0编译时遇到了一个错误,原因是import_array
宏中的一个奇怪的返回语句,我不得不手动删除它才能让它工作,但我在NumPy 1.8.0的GitHub源代码中找不到这个返回语句。