2023: Released '3D-Box-Segment-Anything', combining VoxelNeXt and SAM
2023: 'VoxelNeXt' accepted by CVPR 2023
2022: 'Focal Sparse Conv' accepted by CVPR 2022 as Oral
2022: 'Scale-aware AutoAug' accepted by IEEE T-PAMI
Proposed QeRL: First framework enabling RL training of 32B LLMs on a single H100 80GB GPU, with 1.7× end-to-end speedup and performance matching full fine-tuning