Support for Other Backends

[amirhossein-razlighi]

I believe it is necessary to have support for other backends like CPU, and/or other GPUs (like Apple Silicon). It will allow users to train on a wider variety of devices (although in some cases like CPU, it will be more time-consuming). Right now, gsplat only supports CUDA.

[min-hieu-netropy]

Hi, I think you would be interested in the gaussian-splatting-slang project https://github.com/google/slang-gaussian-rasterization.git which, in theory, could have support for other backends. However, I think most of the code in both project requires a lot of thread group operations that will need to be rewritten as they are only compatible with cuda support at the moment.

[gradeeterna]

Also check out Brush which can train on Apple Silicon and pretty much everything! https://github.com/ArthurBrussee/brush