SS-SFDA : Self-Supervised Source-Free Domain Adaptation for Road Segmentation in Hazardous Environments
This work provides a significant improvement in road segmentation accuracy and training efficiency for autonomous driving systems operating in hazardous environments.
This paper tackles unsupervised road segmentation in adverse weather conditions using a novel self-supervised source-free domain adaptation (SFDA) algorithm. The method outperforms prior works on unsupervised road segmentation and SFDA by at least 10.26% and improves training time by 18-180x, achieving similar mIOU accuracy to prior supervised methods.
We present a novel approach for unsupervised road segmentation in adverse weather conditions such as rain or fog. This includes a new algorithm for source-free domain adaptation (SFDA) using self-supervised learning. Moreover, our approach uses several techniques to address various challenges in SFDA and improve performance, including online generation of pseudo-labels and self-attention as well as use of curriculum learning, entropy minimization and model distillation. We have evaluated the performance on $6$ datasets corresponding to real and synthetic adverse weather conditions. Our method outperforms all prior works on unsupervised road segmentation and SFDA by at least 10.26%, and improves the training time by 18-180x. Moreover, our self-supervised algorithm exhibits similar accuracy performance in terms of mIOU score as compared to prior supervised methods.