RUN.md

Training and Evaluation

After preparing data, run the following scripts to train the model. By default, all models are trained on 8 GPUs.

BEVDepth

Train a distilled BEVDepth with CenterPoint as teacher

./scripts/teacher_to_bevdepth4d/centerpoint2bevdepth.sh

Train a distilled BEVDepth with MVP as teacher

./scripts/teacher_to_bevdepth4d/mvp2bevdepth.sh

BEVFormer

Train a distilled BEVFormer with CenterPoint as teacher

./scripts/teacher_to_bevformer/exp_lidar_r50.sh

Train a distilled BEVFormer with MVP as teacher

./scripts/teacher_to_bevformer/exp_mvp_r50.sh

Evaluation

For single-GPU evaluation, you can run following command

python tools/test.py configs/lidar2camera_bev_distillation/teacher_to_bevformer/lidarformer_to_bevformer_nus_1x1conv_r50.py outputs/lidarformer_to_bevformer_r50/epoch_24.pth --eval mAP