Skip to content

zkyntu/UnLanedet

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

UnLanedet

An advanced lane detection toolbox. UnLanedet contains many advanced lane detection methods to facilitate scientific research and lane detection applications. If you are in China, gitee link may be helpful for you.


What's New

  • [2025-07-21] We release the technical report of UnLanedet: paper_link.
  • [2025-07-18] The latest distributed training code is provided. PR link. Many thanks to the author.
  • [2025-05-30] We support DLA34 and ConvNexT backbone. CLRNet with ConvNext-Tiny gets 80.21 F1 score on CULane.
  • [2025-05-27] We support GSENet and provide the model analysis tools.
  • [2025-05-23] We support GANet, a keypoint-based method, and modulated DCN in mmcv.
  • [2025-05-14] We release v3 version. In this version, we support BezierNet, a parameter-based method.
  • [2025-05-07] We support SRLane, a high-performance model with fast inference speed. Training on the custom dataset is provided in the advanced usage.
  • [2025-04-24] We support distributed training (DDP) and provide the CLRNet-R50 model.
  • [2025-03-12] We release the v2 version. In this version, we add the VIL100 dataset and the ADNet-VIL100 model and provide the fps testing tool. In the future, we will add O2SFormer, keypoint-based methods, and parameter-based methods. Stay tuned.
  • [2025-03-04] We release the Timm library wrapper! Users can directly transfer the advanced backbone to UnLanedet. In the following weeks, we will release the v2 version.
  • [2024-11-10] We release ADNet and LaneATT. Try it!
  • [2024-11-07] We release CondLaneNet and CLRerNet and fix bugs in UnLanedet. Try it!
  • [2024-11-05] We release the v1 version, focusing on 2D lane detection methods.

Installation

See installation instructions.

Getting Started

See Get Started documentation, including the data preparation, the training code, the evaluation code, the resume code, the inference code, and the advanced usage.

Model Zoo and Baselines

We provide a set of lane detection methods. All models and the corresponding weights and the training logs can be found in the Model Zoo.

Advantages of UnLanedet

Compared with other lane detection libraries, e.g., lanedet and PPLanedet, UnLanedet has two obvious advantages: 1) Distributed training is supported. 2) More pretrained models and datasets are provided.

We do not depend on third-party library, such as mmcv series, and all modules and functions can be found in the repo.

License

UnLanedet is released under the Apache 2.0 license.

Contribution

We appreciate all contributions to UnLanedet and welcome pull requests to improve UnLanedet.

Acknowledgement

UnLanedet is built upon detectron2, lanedet and PPLanedet. Many thanks to their great work!

The code of BeizerNet is modified from mmLaneDet. Many thanks to the authors.

Some modules are borrowed from detrex. Many thanks to the authors.

Citing UnLanedet

If you use UnLanedet in your research, please use the following BibTeX entry.

@misc{zhouunlanedet,
  author =       {UnLanedet team},
  title =        {UnLanedet},
  howpublished = {\url{https://github.com/zkyntu/UnLanedet}},
  year =         {2024}
}

About

A unified lane detection toolbox, including CLRNet, ADNet, and CLRerNet.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Contributors 5