DRACO-DehazeNet: An Efficient Image Dehazing Network Combining Detail Recovery and a Novel Contrastive Learning Paradigm

TL;DR

DRACO-DehazeNet is an efficient image dehazing network that combines detail recovery and a novel contrastive learning paradigm, enabling effective dehazing with limited data and improved performance on various haze conditions.

Contribution

The paper introduces a new dehazing network with a dense dilated residual block and a quadruplet loss-based contrastive learning approach for better feature separation.

Findings

Outperforms existing dehazing methods on benchmark datasets.

Effective training with limited data due to contrastive paradigm.

Improved dehazing quality in heavy and non-uniform haze conditions.

Abstract

Image dehazing is crucial for clarifying images obscured by haze or fog, but current learning-based approaches is dependent on large volumes of training data and hence consumed significant computational power. Additionally, their performance is often inadequate under non-uniform or heavy haze. To address these challenges, we developed the Detail Recovery And Contrastive DehazeNet, which facilitates efficient and effective dehazing via a dense dilated inverted residual block and an attention-based detail recovery network that tailors enhancements to specific dehazed scene contexts. A major innovation is its ability to train effectively with limited data, achieved through a novel quadruplet loss-based contrastive dehazing paradigm. This approach distinctly separates hazy and clear image features while also distinguish lower-quality and higher-quality dehazed images obtained from each sub-modules of our network, thereby refining the dehazing process to a larger extent. Extensive tests on a variety of benchmarked haze datasets demonstrated the superiority of our approach. The code repository for this work is available at https://github.com/GreedYLearner1146/DRACO-DehazeNet.