Can Large Pretrained Depth Estimation Models Help With Image Dehazing?

TL;DR

This paper explores how large pretrained depth estimation models can be leveraged to improve image dehazing, demonstrating their generalization and effectiveness across various scenarios.

Contribution

It introduces a plug-and-play RGB-D fusion module that enhances dehazing models using pretrained depth features, showing broad applicability.

Findings

Depth features are consistent across haze levels

The fusion module improves dehazing performance

Approach works across multiple benchmarks

Abstract

Image dehazing remains a challenging problem due to the spatially varying nature of haze in real-world scenes. While existing methods have demonstrated the promise of large-scale pretrained models for image dehazing, their architecture-specific designs hinder adaptability across diverse scenarios with different accuracy and efficiency requirements. In this work, we systematically investigate the generalization capability of pretrained depth representations-learned from millions of diverse images-for image dehazing. Our empirical analysis reveals that the learned deep depth features maintain remarkable consistency across varying haze levels. Building on this insight, we propose a plug-and-play RGB-D fusion module that seamlessly integrates with diverse dehazing architectures. Extensive experiments across multiple benchmarks validate both the effectiveness and broad applicability of our approach.