How to implement high dimensional grid sample?

vision
1

Hi. I want to add a high-dimensional grid sample, e.g., 10D, but I do not know which files should be modified. Any Tips are appreciated.

2

I don’t fully understand the question so could you describe which files you want to modify in the first place?
Do you want to implement a new method into PyTorch?

3

Hi, ptrblck. I want to implement an arbitrary dimensional grid sampler within PyTorch. However, PyTorch only implements a 2D/3D grid sampler. I think some C++/CUDA codes in GridSampler.xxx located in aten/src/ATen/native/cuda should be modified, but I do not know whether other files in other folders should be modified?

4

Yes, the GridSampler.cpp file looks like the correct place to add new functionality.
It also seems you want to add your new methods into the core framework.
If so, you should create a feature request on GitHub and discuss it with the code owners (unless you only want to add it locally).
An easier approach might be to create a custom CUDA extension to provide the new functionality.

5

Thanks for your great help.