Skip to content

Latest commit

 

History

History
112 lines (80 loc) · 4.02 KB

README.md

File metadata and controls

112 lines (80 loc) · 4.02 KB
Raw

ConDA

Project | Paper | Benchmark

ConDA: Unsupervised Domain Adaptation for LiDAR Segmentation via Regularized Domain Concatenation
Lingdong Kong1,*, Niamul Quader2, Venice Erin Liong2
1National University of Singapore, 2Motional
*Work done as an autonomous vehicle intern at Motional

🚘 This is not an official Motional product

About

ConDA aims at processing raw point clouds for unsupervised domain adaptation (UDA) in LiDAR semantic segmentation. It also supports other domain adaptation settings under annotation scarcity, such as semi-supervised domain adaptation (SSDA) and weakly-supervised domain adaptation (WSDA). The main idea of ConDA is to (1) construct an intermediate domain consisting of fine-grained interchange signals from both source and target domains without destabilizing the semantic coherency of objects and background around the ego-vehicle; (2) utilizes the intermediate domain for self-training. Visit our project page to explore more details!

Cross-City UDA Benchmark

Updates

  • [2023.01] - ConDA is accepted to ICRA 2023 🎉!
  • [2022.09] - Our paper is available on arXiv, click here to check it out. Code will be available soon!

Outline

Installation

Please refer to INSTALL.md for the installation details.

Data Preparation

Please refer to DATA_PREPARE.md for the details to prepare the cross-city UDA benchmark with nuScenes,

Getting Started

Please refer to GET_STARTED.md to learn more usage about this codebase.

Main Results

Framework Overview

Domain Discrepancy

Domain Concatenation

TODO List

  • Initial release. 🚀
  • Add license. See here for more details.
  • Add installation details.
  • Add data preparation details.
  • Add evaluation details.
  • Add training details.

License

Creative Commons License
This work is under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

Acknowledgement

We thank Sergi Widjaja, Xiaogang Wang, Dhananjai Sharma, and Edouard Francois Marc Capellier for their insightful reviews and discussions.

Citation

@inproceedings{kong2023conda,
  title = {ConDA: Unsupervised domain adaptation for LiDAR segmentation via regularized domain concatenation},
  author = {Lingdong Kong and Niamul Quader and Venice Erin Liong},
  booktitle = {IEEE International Conference on Robotics and Automation},
  pages = {9338--9345},
  year = {2023}
}

Know More About Motional