[Inference][Win10 pycharm]: Python scripts call C++ code, pass parameter failed.

[zhanxiaopan]

When we try to run cmd "python main.py --cfg ./config/test.yaml" to inference base on the scannet dataset.
An error occured, we found that there may be parameter incompatible issue exist. Could you help to figure it out?

Steps: Compile "torchsparse/backbones/backend/hash/hash_cpu.cpp" and related header files as torch_sparse_c_dll.dll. Then, we excute inference cmd to call C++ API “ts_dll.hash_cpu(coords)”.

Error：ctypes.ArgumentError: argument 1: <class 'TypeError'>: Don't know how to convert parameter 1

Code:
Python: Parameter -》torch.Tensor
from ctypes import *
ts_dll = CDLL("D:/torch_sparse_c_dll/x64/Release/torch_sparse_c_dll.dll")
assert coords.ndim == 2 and coords.shape[1] == 4, coords.shape
coords = coords.cpu()
coords = coords.type(torch.FloatTensor)
return ts_dll.hash_cpu(coords)

C++： Parameter -》at.Tensor
at::Tensor hash_cpu(const at::Tensor idx) {
int N = idx.size(0);
at::Tensor out =
torch::zeros({N}, at::device(idx.device()).dtype(at::ScalarType::Long));
cpu_hash_wrapper(N, idx.data_ptr(), out.data_ptr<int64_t>());
return out;
}

[zhanxiaopan]

Fixed, with the help of Pybind11 tools