A Synthetic-to-Real Dehazing Method based on Domain Unification

2025-09-04

AI Summary

Domain shift between synthetic and real-world hazy images degrades deraining performance, as ideal clean images in existing synthetic datasets often violate the atmospheric scattering model under complex scenes, leading to inconsistent physical relationships across domains. Method: We propose the first domain-unified deraining framework that explicitly models and unifies atmospheric physical priors—specifically, joint constraints on transmission maps and global atmospheric light—across synthetic and real domains. Our approach integrates a deep neural network with a differentiable physics layer, enabling physics-consistent domain adaptation without paired real-world ground-truth data. Results: Extensive experiments demonstrate significant improvements over state-of-the-art methods on multiple real-world hazy benchmarks. Both quantitative metrics and qualitative visual results confirm the robustness, generalizability, and superiority of our method in handling domain discrepancies while preserving physical plausibility.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Application Domains: Natural Sciences

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Security and Privacy: Data transparency and provenance
• Systems and Infrastructure for Web, Mobile and WoT: Web applications in cross-disciplinary domains and verticals such as mixed reality, smart cities, and digital health

Abstract

Due to distribution shift, the performance of deep learning-based method for image dehazing is adversely affected when applied to real-world hazy images. In this paper, we find that such deviation in dehazing task between real and synthetic domains may come from the imperfect collection of clean data. Owing to the complexity of the scene and the effect of depth, the collected clean data cannot strictly meet the ideal conditions, which makes the atmospheric physics model in the real domain inconsistent with that in the synthetic domain. For this reason, we come up with a synthetic-to-real dehazing method based on domain unification, which attempts to unify the relationship between the real and synthetic domain, thus to let the dehazing model more in line with the actual situation. Extensive experiments qualitatively and quantitatively demonstrate that the proposed dehazing method significantly outperforms state-of-the-art methods on real-world images.

Problem

• Addressing synthetic-to-real domain shift in image dehazing
• Correcting imperfect clean data collection in real domains
• Unifying atmospheric physics models across synthetic and real domains

Innovation

• Domain unification bridges synthetic and real dehazing
• Addresses imperfect clean data collection issues
• Enhances model alignment with real atmospheric physics