Official code for Distribution-Aware Continual Test-Time Adaptation for Semantic Segmentation.
This repo contains detailed implementation for teacher-student forward propagation, uncertainty calculation, pixel-level distribution shift evaluation and parameter selection and accumulation.
Training process can refer to cotta paper.
Experimental results of DAT model on Cityscapes_to_ACDC and SHIFT datasets can be found in acdc and shift.
Please cite our work if you find it useful.
@misc{ni2024distributionaware,
title={Distribution-Aware Continual Test-Time Adaptation for Semantic Segmentation},
author={Jiayi Ni and Senqiao Yang and Ran Xu and Jiaming Liu and Xiaoqi Li and Wenyu Jiao and Zehui Chen and Yi Liu and Shanghang Zhang},
year={2024}
}