Does opencv cuda plan to support torch.Tensor or tritonserver.Tensor?

[timeuser4]

In my project, I encountered some pre-processing using the cpu version of opencv (such as lanzcous4 interpolation resize). I manually wrote them into cuda operators as torch.Tensor and triton.Tensor, but I think if opencv can seamlessly support these two formats, there will be no need to do this step, and it will be much faster.

[cudawarped]

@timeuser4 If I understand your question correctly, you want to perform actions which are not supported by the OpenCV CUDA modules. You have chosen to perform this in pytorch or triton so the processing is performed on the GPU but you need a way to pass this into OpenCV. Is that correct?

If so and you can get the CUDA pointer from torch.Tensor you may be able to use createGpuMatFromCudaMemory().

[timeuser4]

Yes, you understand exactly what I mean, thank you very much. In addition,If I want to contribute to OpenCV, what else should I prepare besides downloading the code, writing the code, compiling and running the code, and providing the time required to run the code? I want to complete the interpolation method that is not available in opencv::cuda::resize.I'm a complete novice when it comes to contributing on github, so I'm sorry if I bother you.