How to pass opencv Mat type to Python and get an array back?

Question

I have a deep learning python code which gets an image and returns an array of detection positions in that image.

I have developed all other parts of my project in C++ and I want to pass an opencv Mat image (cv::Mat) to python script and it would return an array to C++.

Is there a way to pass these data types from/to C++/Python? what funtions should I use?

if it is not performance critical, the easiest is to store it in a file and read it again — 463035818_is_not_an_ai
– 463035818_is_not_an_ai, Commented Mar 5, 2019 at 11:42
Is this what you are looking for? Just pass a pointer to the data to your python script and unpack it as struct. Then pack it again to pass it back to your C++ application. Or compile your python script down to C with Cython and try to integrate that into your code. Another way to do it would be to entirely do it in C++ or Python and not mix it, if performance is really critical. Tensorflow has a C++ API. — jaaq
– jaaq, Commented Mar 5, 2019 at 12:23
"Is there a way to pass these data types from/to C++/Python?" -- Of course, have a look at how the OpenCV Python bindings work. There are multiple posts on SO about how to handle mapping Mat to ndarray. There's also Boost.Numpy. — Dan Mašek
– Dan Mašek, Commented Mar 5, 2019 at 13:10

Vaibhav gusain · Accepted Answer · 2019-03-05 12:26:48Z

1

So this can be achieved in 2-ways.

1.) you can develop a backend server for python and then expose those api's , and then you can use your api's to send the data and get the results back.

2.) also you can make your python function called by a c++ program ,for that you can use lib like Python.h to call and receive data from the python script.

Example lets say you have a python file named aiDosomething.py which has a function detect which do the detection task. then in c++ you can do something like

#include <Python.h>

int main()
{

pModule = PyImport_ImportModule("aiDosomething");

//chose the function in file Sample

pFunc = PyObject_GetAttrString(pModule,"detect");

pDict = PyModule_GetDict(pModule);   // Create a dictionary for the contents of the module.

//make the args to be passes in the function

pArgs = PyTuple_New(1);     // Create a Python tuple to hold the arguments to the method.
PyTuple_SetItem(pArgs, 0, insertArray); // insertArray can be your cv::mat array

//run the code and get the result
PyObject* pResult = PyObject_CallObject(pFunc, pArgs);
return 0;

}

for keeping things simple you can put your python code and c++ under same directory.

answered Mar 5, 2019 at 12:26

Vaibhav gusain

4163 silver badges9 bronze badges

Sign up to request clarification or add additional context in comments.

6 Comments

MJay Over a year ago

This is the error returned when I want to use cv::Mat as insertArray: "no suitable conversion from "cv::Mat" to "PyObject *" exists"

Vaibhav gusain Over a year ago

try converting cv::mat array to normal array and send it

MJay Over a year ago

how is that possible? Just type casting?

Vaibhav gusain Over a year ago

you can think of the image as an 2D array,you first initialize the 2d array having shape equal to cv::Mat, then you can simply iterate through every value in the cv::mat and then setting the value to the normal c++ array. Also if you dont do much in c++ you can directly pass your imagePath to python and load the image in the function itself.

MJay Over a year ago

I did it with no error by PyTuple_SetItem(pArgs, 0, (PyObject*)(char*)img.data); but nothing is shown in python script. Even print(img.shape()) does nothing!

|

Collectives™ on Stack Overflow

How to pass opencv Mat type to Python and get an array back?

1 Answer 1

6 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

6 Comments

Your Answer

Sign up or log in

Post as a guest

Related