Can we implement depth-wise convolution efficiently with ME?

NVIDIA / MinkowskiEngine

Minkowski Engine is an auto-diff neural network library for high-dimensional sparse tensors

https://nvidia.github.io/MinkowskiEngine

Other

2.43k stars 360 forks source link

Can we implement depth-wise convolution efficiently with ME? #500

Open Labo9527 opened 1 year ago

Labo9527 commented 1 year ago

depth-wise convolution is popular now and it requires group convolution. However, it seems like the ME does not support group convolution and we must implement it by ourselves, which makes it slow. I wonder if there is a plan to support group convolution?