Improved Techniques for Learning to Dehaze and Beyond: A Collective Study

TL;DR

This paper presents improved techniques for single image dehazing and high-level visual understanding of hazy images, demonstrating significant performance gains through novel loss functions and advanced detection modules.

Contribution

It introduces perception-driven loss functions for dehazing and domain-adaptive detection modules for better understanding of hazy images, advancing the state-of-the-art in both tasks.

Findings

Perception-driven loss enhances dehazing quality

Advanced modules improve detection in hazy conditions

Proposed methods outperform existing approaches

Abstract

Here we explore two related but important tasks based on the recently released REalistic Single Image DEhazing (RESIDE) benchmark dataset: (i) single image dehazing as a low-level image restoration problem; and (ii) high-level visual understanding (e.g., object detection) of hazy images. For the first task, we investigated a variety of loss functions and show that perception-driven loss significantly improves dehazing performance. In the second task, we provide multiple solutions including using advanced modules in the dehazing-detection cascade and domain-adaptive object detectors. In both tasks, our proposed solutions significantly improve performance. GitHub repository URL is: https://github.com/guanlongzhao/dehaze