Official implementation of our AAAI paper: Self-supervised Likelihood Estimation with Energy Guidance for Anomaly Segmentation in Urban Scenes
Self-supervised Likelihood Estimation with Energy Guidance for Anomaly Segmentation in Urban Scenes (AAAI 2024)
This is the pytorch implementation of the paper (accepted by AAAI 2024).
For the training set and test set, please refer to link to download cityscapes and fishyscapes datasets. Place the downloaded dataset at ./data/.
The docker file for running our SLEEG can be found in ./docker, where the submission file of our SLEEG on the Fishyscapes dataset is also available.
bash scripts/train.sh
bash scripts/static.sh
The trained weights of SLEEG are available at link.
Our results on the leaderboard of Fishyscapes can be found in link, where anonymous submission (mall) is our SLEEG.
The code is built on mmseg. Many thanks for their great work. If you find this repo useful for your research, please consider citing our paper:
@misc{tu2023selfsupervised,
title={Self-supervised Likelihood Estimation with Energy Guidance for Anomaly Segmentation in Urban Scenes},
author={Yuanpeng Tu and Yuxi Li and Boshen Zhang and Liang Liu and Jiangning Zhang and Yabiao Wang and Chengjie Wang and Cai Rong Zhao},
year={2023},
eprint={2302.06815},
archivePrefix={arXiv},
primaryClass={cs.CV}
}