This codebase was tested with the following environment configurations. It may work with other versions.

• CUDA 11.3/CUDA 11.7
• Python 3.8+
• PyTorch 1.9.0/PyTorch 1.13.0
• MMCV 1.6.0
• MMDetection 2.26.0
• MMSegmentation 0.29.1
• MMRotate 0.3.4
• MMDetection3d 1.0.0rc3

Please refer to getting_started.md for installation.

We use KITTI and nuScenes datasets, please follow the official instructions for set up (KITTI instruction, nuScenes instruction).

Please make sure you have set up the environments and you can start knowledge distillation by running

DEVICE_ID = {gpu_id} # for single gpu CUDA_VISIBLE_DEVICES=$DEVICE_ID python tools/train_kd.py {distillation_cfg} # for multiple gpus bash ./tools/dist_train_kd.sh <distillation_cfg> 8 

Many thanks to following codes that help us a lot in building this codebase:

• PointDistiller
• Towards Efficient 3D Object Detection with Knowledge Distillation
• mmdetection
• mmdetection3d

If you find our work useful in your research, please consider citing:

To be continued