xinge008 / Cylinder3D

Rank 1st in the leaderboard of SemanticKITTI semantic segmentation (both single-scan and multi-scan) (Nov. 2020) (CVPR2021 Oral)
Apache License 2.0
858 stars 180 forks source link

Regarding other sparse conv libraries #138

Closed akshay-antony closed 2 years ago

akshay-antony commented 2 years ago

Hi, Thanks for the amazing work and concise explanation in the paper. But can I use other sparse Conv libraries like ME, or torch sparse or is there any reason for using spconv. And is spconv tensorrt compatible?

xinge008 commented 2 years ago

We follow the popular setting in 3D detection (like SECOND, PointPillar) to use the spconv as the basic block; Other sparse convs may work too, while we do not have a try.