search

NVIDIA / MinkowskiEngine

Minkowski Engine is an auto-diff neural network library for high-dimensional sparse tensors
https://nvidia.github.io/MinkowskiEngine
Other
2.48k stars 367 forks source link

Voxelization of input #375

Closed shreyakatari closed 3 years ago

shreyakatari commented 3 years ago

Hi @chrischoy Can we give voxelized input to training instead of original input? Will it affect the performance of training as their is prevoxel transform as well in semantic segmentation. Thank you.

chrischoy commented 3 years ago

You can give either the original or prevoxelized input. If you have more sophisticated network with interpolation, there could be some performance gain but not too much different if the voxel size is really small.