sj-li / MINet

Multi-scale Interaction for Real-time LiDAR Data Segmentation on an Embedded Platform (RA-L)
40 stars 6 forks source link
lidar semantic-segmentation

Multi-scale Interaction for Real-time LiDAR Data Segmentation on an Embedded Platform (RA-L)

Dependence:

  1. Accoding to LiDAR-Bonnetal (https://github.com/PRBonn/lidar-bonnetal/tree/master/train)
  2. flops-counter.pytorch (https://github.com/sovrasov/flops-counter.pytorch)
  3. Edge files: Edges

Infer:

  1. put 'sequences' folder under 'data/'
  2. 'python infer.py --dataset data --arch_cfg config/arch/config_file --data_cfg config/labels/semantic-kitti.yaml --checkpoint checkpoints/checkpoint_file --log predictions'

Attention

  1. Only infer validation set, refer 'lib/user.py' line 70-80.
  2. pay attention to kNN setting. (In 'config/arch/*.yaml')

Citation

Please cite the following paper if you use this repository in your reseach.


@ARTICLE{9633188,
  author={Li, Shijie and Chen, Xieyuanli and Liu, Yun and Dai, Dengxin and Stachniss, Cyrill and Gall, Juergen},
  journal={IEEE Robotics and Automation Letters}, 
  title={Multi-Scale Interaction for Real-Time LiDAR Data Segmentation on an Embedded Platform}, 
  year={2022},
  volume={7},
  number={2},
  pages={738-745},
  doi={10.1109/LRA.2021.3132059}}