19.[VoTr](https://github.com/PointsCoder/VOTR) implements submanifold [voxel transformers](https://openaccess.thecvf.com/content/ICCV2021/papers/Mao_Voxel_Transformer_for_3D_Object_Detection_ICCV_2021_paper.pdf) using [SpConv](https://github.com/traveller59/spconv).
19.[VoTr](https://github.com/PointsCoder/VOTR) implements submanifold [voxel transformers](https://openaccess.thecvf.com/content/ICCV2021/papers/Mao_Voxel_Transformer_for_3D_Object_Detection_ICCV_2021_paper.pdf) using [SpConv](https://github.com/traveller59/spconv).
20.[Mix3D](https://github.com/kumuji/mix3d) brings [MixUp](https://openreview.net/forum?id=r1Ddp1-Rb) to the sparse setting—0.781 average IOU for ScanNet 3D semantic segmentation.
20.[Mix3D](https://github.com/kumuji/mix3d) brings [MixUp](https://openreview.net/forum?id=r1Ddp1-Rb) to the sparse setting— 0.781 average IOU for ScanNet 3D semantic segmentation.
21.[Point Transformer V3](https://arxiv.org/abs/2312.10035) uses sparse convolutions as an enhanced conditional positional encoding (xCPE); 0.794 average IOU for ScanNet 3D semantic segmentation.