process exited with code -1073741819 (0xC0000005) after calling function in pyd created by pybind11

[zhusihan-python]

hello, my main GUI process crash after calling function in pyd created by pybind11, the function wrapped by it executed successfully, yet something wrong happened when the pyd exit. the 0xC0000005 indicated memory issues right?

my code is as below:

#include "itkImageFileReader.h"
#include "itkTileConfiguration.h"
#include "itkTileMontage.h"

#include <pybind11.h>
#include <stl.h>
#include <numpy.h>
#include "itkImage.h"
#include "itkImportImageFilter.h"
#include <Python.h>
#include <vector>

using namespace std;
namespace py = pybind11;
using ImageReturnType = itk::Image<unsigned char,2>;

#define T unsigned char
#define N 2
ImageReturnType::Pointer
numpyToImage(py::array_t<T> & arr)
{
  py::buffer_info buf = arr.request();
  T *             rawDataPtr = (T *)buf.ptr;

  // Instantiate an itk image object
  typedef itk::Image<T, N>             ImageType;
  typedef itk::ImportImageFilter<T, N> ImportFilterType;
  auto                                 importFilter = ImportFilterType::New();

  // Define the size and index of the image region
  ImportFilterType::SizeType size;
  size[0] = buf.shape[1];
  size[1] = buf.shape[0];

  ImportFilterType::IndexType start;
  start.Fill(0);

  ImportFilterType::RegionType region;
  region.SetIndex(start);
  region.SetSize(size);

  importFilter->SetRegion(region);

  const itk::SpacePrecisionType origin[N] = { 0.0, 0.0 };
  importFilter->SetOrigin(origin);
  const itk::SpacePrecisionType spacing[N] = { 1.0, 1.0 };
  importFilter->SetSpacing(spacing);

  const unsigned int numberOfPixels = size[0] * size[1];
  importFilter->SetImportPointer(rawDataPtr, numberOfPixels, true);
  importFilter->Update();

  return importFilter->GetOutput();
}

class GILManager
{
  PyGILState_STATE gstate;

public:
  GILManager()
    : gstate(PyGILState_Ensure())
  {}

  ~GILManager() { PyGILState_Release(gstate); }
};
#define PYTHON_GIL GILManager __gil_manager

template <unsigned Dimension, typename PixelType, typename AccumulatePixelType>
int
refineMontage(std::string inFile, std::string outFile, std::vector<py::array_t<T>> imageList)
{
  try
  {
    PYTHON_GIL;
    itk::TileConfiguration<Dimension> stageTiles;
    stageTiles.Parse(inFile);
    itk::TileConfiguration<Dimension> actualTiles = stageTiles; // result - only positions need to be updated

    using TileConfig = itk::TileConfiguration<Dimension>;
    using TransformType = itk::TranslationTransform<double, Dimension>;
    using ScalarImageType = itk::Image<PixelType, Dimension>;
    typename ScalarImageType::SpacingType sp;
    sp.Fill(1.0);

    // instantiate and invoke the montaging class
    using MontageType = itk::TileMontage<ScalarImageType>;
    typename MontageType::Pointer montage = MontageType::New();
    montage->SetMontageSize(stageTiles.AxisSizes);

    for (size_t t = 0; t < stageTiles.LinearSize(); t++)
    {
      typename TileConfig::PointType    origin = stageTiles.Tiles[t].Position;
      typename ScalarImageType::Pointer image = numpyToImage((py::array_t<T>)imageList[t]);

      // tile configurations are in pixel (index) coordinates
      // so we convert them into physical ones
      sp = image->GetSpacing();
      for (unsigned d = 0; d < Dimension; d++)
      {
        origin[d] *= sp[d];
      }
      image->SetOrigin(origin);
      image->Update();

      montage->SetInputTile(t, image);
    }
    montage->Update(); // calculate registration transforms
    for (size_t t = 0; t < actualTiles.LinearSize(); t++)
    {
      typename MontageType::TileIndexType  ind = actualTiles.LinearIndexToNDIndex(t);
      const TransformType *                regTr = montage->GetOutputTransform(ind);
      const itk::Vector<double, Dimension> regPos = regTr->GetOffset();
      // transform physical into index shift and modify original position
      for (unsigned d = 0; d < Dimension; d++)
      {
        actualTiles.Tiles[t].Position[d] = stageTiles.Tiles[t].Position[d] - regPos[d] / sp[d];
      }
    }
    actualTiles.Write(outFile);
    return EXIT_SUCCESS;
  }
  catch (itk::ExceptionObject & exc)
  {
    std::cerr << exc;
  }
  catch (std::runtime_error & exc)
  {
    std::cerr << exc.what();
  }
  catch (...)
  {
    std::cerr << "Unknown error has occurred" << std::endl;
  }
  return EXIT_FAILURE;
}

PYBIND11_MODULE(RefineMontage, m)
{
  m.def("NewRefineMontage", &refineMontage<2, unsigned char, double>, "refineMontage ready to use!>");
}

code in python side like this:

from RefineMontage import NewRefineMontage

imgs = []  # numpy.array list
res = NewRefineMontage(src, dst, imgs)
print("res", res)

i got my result.txt file, and the return code is success(0), no exceptions are caught, but it crashed when exit

how can i fix this error, or is there an elegant way to exit the pyd library created by pybind11

environments: windows 11 python 3.10.11 pybind11 2.10.4, i build the pyd file with VS 2022

[zhusihan-python]

i build a debug version pyd file, and got an error

File: minkernel\crts\ucrt\src\appcrt\heap\debug_heap.cpp
Line: 904

Expression: _CrtlsValidHeapPointer(block)
does this related to the exit issuse

[zhusihan-python]

i found that the elements in imageList should be release after montage->Update()

for (auto & arr: imageList) 
{
  arr.release();
}
imageList.clear();
imageList.shrink_to_fit();

now the return code of call of refineMontage one time is 0, yet it still got -1073741819 (0xC0000005) after around 20 times calling

[zhusihan-python]

sometimes the error code is -1073741676 (0xC0000094)