data_preparation.md

You can download nuScenes 3D detection data HERE and unzip all zip files. You can only download v1.0-trainval version conveniently. Like the general way to prepare dataset, it is recommended to symlink the dataset root to $CoP/data.

We typically need to organize the useful data information with a .pkl. To prepare these files for nuScenes, run the following command:

python tools/create_data.py  
python tools/data_converter/prepare_nuscenes_for_bevdet.py  

Then download nuscenes panoptic dataset HERE. Symlink the dataset root to $CoP/data/nuscenes and replace .json file following HERE. Run the following command to process point cloud with panoptic labels and prepare for 3D occupancy Ground Truth generation.

python tools/data_converter/prepare_processedPoints.py

Next, run the following command to generate ldo-occ ground truths.

python ./data_gen/ldo_data_gen.py