DAMix: A Density-Aware Mixup Augmentation for Single Image Dehazing under Domain Shift

TL;DR

This paper introduces DAMix, a density-aware augmentation technique for single image dehazing that reduces domain gaps and improves adaptation across datasets, especially under limited data conditions.

Contribution

DAMix is a novel augmentation method that minimizes Wasserstein distance to target domain hazy images, enhancing domain adaptation and data efficiency in dehazing models.

Findings

DAMix reduces domain gap and improves dehazing performance across datasets.

Training with DAMix on half the data outperforms full-data training without DAMix.

DAMix complies with atmospheric scattering model principles.

Abstract

Deep learning-based methods have achieved considerable success on single image dehazing in recent years. However, these methods are often subject to performance degradation when domain shifts are confronted. Specifically, haze density gaps exist among the existing datasets, often resulting in poor performance when these methods are tested across datasets. To address this issue, we propose a density-aware mixup augmentation (DAMix). DAMix generates samples in an attempt to minimize the Wasserstein distance with the hazy images in the target domain. These DAMix-ed samples not only mitigate domain gaps but are also proven to comply with the atmospheric scattering model. Thus, DAMix achieves comprehensive improvements on domain adaptation. Furthermore, we show that DAMix is helpful with respect to data efficiency. Specifically, a network trained with half of the source dataset using DAMix can achieve even better adaptivity than that trained with the whole source dataset but without DAMix.