FogAdapt: Self-Supervised Domain Adaptation for Semantic Segmentation of Foggy Images
(Published at Neurocomputing 2022)
Javed Iqbal, Rehan Hafiz, Mohsen Ali
Information Technology University, Pakistan
Resizing the foggy input image has a different effect on the self-entropy (SE) map. Segmenting foggy scenes at a higher scale provides extra local context, i.e., minimizes the effect of fog by producing better segmentation with comparatively sharp edges. Contrary to that, segmenting images at a lower scale produce better outputs for large and near to camera objects disguised by fog.
Abstract
This paper presents FogAdapt, a novel approach for domain adaptation of semantic segmentation for dense foggy scenes. Large variations in the visibility of the scene due to weather conditions, such as fog, smog, and haze, exacerbate the domain shift, thus making unsupervised adaptation in such scenarios challenging. We propose a self-entropy and multi-scale information augmented self-supervised domain adaptation method (FogAdapt) to minimize the domain shift in foggy scenes segmentation. Our experiments demonstrate that FogAdapt significantly outperforms the current state-of-the-art in semantic segmentation of foggy images. Specifically, by considering the standard settings compared to state-of-the-art (SOTA) methods, FogAdapt gains 3.8% on Foggy Zurich, 6.0% on Foggy Driving-dense, and 3.6% on Foggy Driving in mIoU when adapted from Cityscapes to Foggy Zurich.
RESULTS
Pdf, Code and Results
Click on image to view it
Authors' Information
Javed Iqbal: PhD Student, Intelligent Machines Lab, ITU, Lahore, Pakistan
Email: javed.iqbal@itu.edu.pk
Web: linkedin
Rehan Hafiz: Professor, Dept of Computer Engineering, ITU Pakistan
Email: rehan.hafiz@itu.edu.pk
Web: https://im.itu.edu.pk/
Yu-TsehChi, ——
Email: jchi@fb.com
Web: —-
Dr. Mohsen Ali, Assistant Professor, Intelligent Machines Lab, ITU, Lahore, Pakistan
Email: mohsen.ali@itu.edu.pk
Web: https://im.itu.edu.pk/
Citation
Iqbal, Javed, Rehan Hafiz, and Mohsen Ali. “FogAdapt: Self-Supervised Domain Adaptation for Semantic Segmentation of Foggy Images.” Neurocomputing (2022).