How to express this"a[0:3, 0:3]" in python as c++(pybind11)? - pybind11

Question

Recently, when using pybind11, I encountered how to use python code a[0: 3,0: 3] to achieve this problem. There is currently a slice function, but I did not find the relevant user manual, resulting in the incorrect access.
ps: I have tried this way you said before, but when I printed out the matrix, I found it was not right. I don’t know the reason. Please help me. Thank you very much.
cout result
py::scoped_interpreter guard{};
py::module np = py::module::import("numpy");
py::object random = np.attr("random");
py::module sys = py::module::import("sys");
py::print(sys.attr("path"));
py::module scipy = py::module::import("scipy.ndimage");
// get scipy.optimize.curve_fit
py::function affine_transform = scipy.attr("affine_transform");
py::array_t<float> new_affine = np.attr("eye")(4);
py::array_t<float> new_af = new_affine[py::make_tuple(py::slice(0, 3, 1), py::slice(0, 3, 1))];
std::cout << numpy_to_cv_mat(new_affine) << endl;
std::cout << new_af.size() << endl;
std::cout << numpy_to_cv_mat(new_af) << endl;
cv::Mat numpy_to_cv_mat(py::array_t<float>& input) { py::buffer_info buf = input.request(); cv::Mat mat(buf.shape[0], buf.shape[1], CV_32FC1, (float*)buf.ptr); return mat; }

Answer 1

There are a few details required need here. First, py::slice(start, stop, step) creates a slice object via pybind11, like what would be created by slice(start, stop, step) in Python.
Second, given a py::array object a, the [] operator does work for slicing in C++ (a[py::slice(s0,s1,st)]) but there is a big caveat: a[] allows (and compiles) with multiple arguments, but only one argument is actually used for slicing, so a[slice(...), slice(...)] only applies the slice on the first dimension.
For multi-dimensional slicing, the [] operator must be passed a py::tuple of py::slice objects. For example, a[0:3,0:3] in Python would be translated to the following in C++:
// a is py::array
a[py::make_tuple(py::slice(0,3,1), py::slice(0,3,1))]
Putting this together, here's a full example which creates and slices a 2D array based on start/stop inputs:
#include <pybind11/pybind11.h>
#include <pybind11/numpy.h>
namespace py = pybind11;
py::array do_slice(py::array a, py::int_ start, py::int_ stop) {
auto res = a[py::make_tuple(py::slice(start, stop, 1), py::slice(start, stop, 1))];
return res;
}
PYBIND11_MODULE(ex, m) {
m.def("do_slice", &do_slice);
}
Some usage examples after compiling:
>>> import numpy as np
>>> a = np.arange(16).reshape(4,4)
>>> a
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11],
[12, 13, 14, 15]])
>>> import ex
>>> ex.do_slice(a, 0, 1)
array([[0]])
>>> ex.do_slice(a, 0, 2)
array([[0, 1],
[4, 5]])
>>> ex.do_slice(a, 0, 3)
array([[ 0, 1, 2],
[ 4, 5, 6],
[ 8, 9, 10]])