- Dataset Preparation
- Check and follow the instruction in
scripts/prepare_nuscenes.sh to set related variables properly and run the script to prepare the nuScenes dataset.
- It will take a long time (a few hours) to download and process the dataset.
bash scripts/prepare_nuscenes.sh
- Model Training & Evaluation
- Set the config file, working directory, GPU numbers etc. in
scripts/run_pointpillars.sh, then run it.
- Note: our optimization for voxelization has been integrated into docker to accelerate the performance of training and inference.
bash scripts/run_pointpillars.sh