Home

Awesome

arXiv

FPS-Net

Code for "FPS-Net: A convolutional fusion network for large-scale LiDAR point cloud segmentation", accepted by ISPRS journal of Photogrammetry and Remote Sensing
By Aoran Xiao, Xiaofei Yang, Shijian Lu, Dayan Guan, Jiaxing Huang

Full Paper

Install

conda create -n FPSNet python=3.7
source activate FPSNet
cd /ROOT/
pip install -r requirements.txt

Dataset

Download SemanticKITTI dataset from official website The dataset structure should be

./
├── 
├── ...
└── path_to_data_shown_in_config/
      └── sequences
            ├── 00/           
            │   ├── velodyne/	
            |   |	   ├── 000000.bin
            |   |	   ├── 000001.bin
            |   |	   └── ...
            │   ├── labels/ 
            |   |      ├── 000000.label
            |   |      ├── 000001.label
            |   |      └── ...
            |   ├── calib.txt
            |   ├── poses.txt
            |   └── times.txt
            └── 08/

Train

Revise dataset path in train.sh and run

cd /train/tasks/semantic
sh train.sh

Inference and Test

Revise dataset path in test.sh and run

cd /train/tasks/semantic
sh test.sh

We provide pre-trained weights, you can download and check (SemanticKITTI: mIoU=57.1 over the testing set; 59.1 over the validation set).

Citation

If you use this code, please cite:

@article{xiao2021fps,
  title={FPS-Net: A convolutional fusion network for large-scale LiDAR point cloud segmentation},
  author={Xiao, Aoran and Yang, Xiaofei and Lu, Shijian and Guan, Dayan and Huang, Jiaxing},
  journal={ISPRS Journal of Photogrammetry and Remote Sensing},
  volume={176},
  pages={237--249},
  year={2021},
  publisher={Elsevier}
}

Acknowledgement

Part of code is borrowed from lidar-bonnetal, thanks for their sharing!

Related Repos