NFC_relocalization

contributed by Kyeongsu Kang, Minjae Lee, Hyeonwoo Yu at the UNIST.

This is a prototype version and will be updated to a more user-friendly version for future execution.

NFC_relocalization is Lidar-base Large-scale Global Place Recognition and Relocalization method by estimate 6-DOF transform

Enviroment setup

NFC_relocalization is based on LCDnet. If you want to use NFC_relocalization, you can use LCDnet's docker or this way

1. Install PyTorch (make sure to select the correct cuda version)
2. Install the requirements pip install -r requirements.txt
3. Install spconv <= 2.1.25 (make sure to select the correct cuda version, for example pip install spconv-cu113==2.1.25 for cuda 11.3)
4. Install OpenPCDet
5. Install faiss-cpu - NOTE: avoid installing faiss via pip, use the conda version, or build it from source alternatively.

Preprocessing

Download the KITTI dataset and Preprocessing for your training:

python -m data_process.generate_loop_GT_KITTI --root_folder KITTI_ROOT

KITTI_ROOT is your download place the KITTI dataset

Ground Plane Removal

You can use Ground Plane Removal remove_ground_plane_kitti.py in data_process. However, for better results, we recommend using patchwork++.

Training

The training script is not support to parallel GPU learning. During training, you can only use one GPU.

To train on the KITTI dataset:

python -m training_KITTI_DDP --root_folder KITTI_ROOT --dataset kitti --batch_size B --without_ground

Acknowledgements

This implementation is based on LCDnet