fix sparse_quantize to work with int coordinates

NVIDIA / MinkowskiEngine

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

https://nvidia.github.io/MinkowskiEngine

Other

2.47k stars 367 forks source link

fix sparse_quantize to work with int coordinates #508

Open soCzech opened 1 year ago

soCzech commented 1 year ago

Do not call torch.floor on int tensors as it returns RuntimeError: "floor_cuda" not implemented for 'Int' or RuntimeError: "floor" "_vml_cpu" not implemented for 'Int'.