SuperLine3D: Self-supervised Line Segmentation and Description for LiDAR Point Cloud
Xiangrui Zhao, Sheng Yang, Tianxin Huang, Jun Chen, Teng Ma, Mingyang Li and Yong Liu
ECCV 2022 Poster
-
Code for training on synthetic and real data, and the synthetic data preparation scripts (2022-7-21).
-
Evaluation code and metrics (expected 2022-8-21).
Our code was tested using Tensorflow 1.15.5 and Python 3.8. To use the code, you need to first install the following python packages:
pip install tensorflow-gpu==1.15.5 h5py open3d tqdm visdom multiprocessing Geometry3D
pip install git+https://github.com/utiasSTARS/liegroups.git# Generate synthetic data
cd scripts
python generate_synthetic_pc.py synthetic_data/
# Train on synthetic data
cd ..
python train_synthetic_data.py --load_folder scripts/synthetic_data/line_pole_noise/ Download labeled Lidar data from Google Drive
Change batch_size according to your VRAM. It must be even.
python train_superline3d.py --loader_folder <path to your train data>/kitti_reg_diff35/Download test Lidar data from Google Drive
python inference_superline3d.py --load_folder <path to your data>/kitti_reg_test_diff5_2w5/ --pred_path <save path> --best_model xxx.ckpt @inproceedings{zhao2022superline3d,
title={SuperLine3D: Self-supervised Line Segmentation and Description for LiDAR Point Cloud},
author={Zhao, Xiangrui and Yang, Sheng and Huang, Tianxin and Chen, Jun and Ma, Teng and Li, Mingyang and Liu, Yong},
booktitle={European Conference on Computer Vision},
year={2022},
organization={Springer}
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent