Please ensure you have prepared the environment and the SemanticKITTI dataset.

Train and Test

Stage-1: Class-Agnostic Query Proposal

Train QPN with 4 GPUs

./tools/dist_train.sh ./projects/configs/voxformer/qpn.py 4

Eval QPN with 4 GPUs

./tools/dist_test.sh ./projects/configs/voxformer/qpn.py ./path/to/ckpts.pth 4

Stage-2: Class-Specific Voxel Segmentation

Train VoxFormer with temporal information with 4 GPUs

./tools/dist_train.sh ./projects/configs/voxformer/voxformer-T.py 4

Eval VoxFormer with temporal information with 4 GPUs

./tools/dist_test.sh ./projects/configs/voxformer/voxformer-T.py ./path/to/ckpts.pth 4