DiffLoc

DiffLoc: Diffusion Model for Outdoor LiDAR Localization

Visualization

Environment

python 3.9
pytorch 1.13
cuda 11.6

source install.sh

Dataset

We support the Oxford Radar RobotCar and NCLT datasets right now.

The data of the Oxford and NCLT dataset should be organized as follows:

data_root
├── 2019-01-11-14-02-26-radar-oxford-10k
│   ├── xxx.bin
│   ├── xxx.bin
├── Oxford_pose_stats.txt
├── train_split.txt
├── valid_split.txt

Data prepare

NCLT: We use NCLT Sample Python Scripts to preprocess velodyne_sync to speed up data reading. We provided within it nclt_precess.py.
Oxford&NCLT: We use SPVNAS to generate static object masks to train the SOAP module. You need to download the code for SPVNAS and run the data_prepare.py.

Run

Download the pretrained ViT model

We initialize DiffLoc's feature learner with DINOv2.

Train

accelerate launch --multi_gpu --num_processes 4 --mixed_precision fp16 train.py

Test

python test.py

Model zoo

The models of DiffLoc on Oxford, and NCLT can be downloaded here.

Acknowledgement

We appreciate the code of RangeVit and PoseDiffusion they shared.

Citation

@inproceedings{li2024diffloc,
  title={DiffLoc: Diffusion Model for Outdoor LiDAR Localization},
  author={Li, Wen and Yang, Yuyang and Yu, Shangshu and Hu, Guosheng and Wen, Chenglu and Cheng, Ming and Wang, Cheng},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={15045--15054},
  year={2024}
}

liw95/DiffLoc