IDDM: Bridging Synthetic-to-Real Domain Gap from Physics-Guided Diffusion for Real-world Image Dehazing

2025-04-30

AI Summary

Existing dehazing models trained solely on synthetic data suffer from poor generalization to real-world hazy scenes due to domain shift between synthetic and real haze distributions. Method: We propose the Physics-Guided Image Dehazing Diffusion Model (IDDM), the first diffusion-based dehazing framework that explicitly incorporates the atmospheric scattering model into the diffusion process. IDDM introduces a joint haze-noise forward corruption mechanism and a decoupled reverse sampling strategy, enabling robust real-scene dehazing using only synthetic training data. Built upon a conditional DDPM framework, it employs a U-Net backbone, physics-driven noise scheduling, and atmospheric scattering constraints in forward modeling. Results: IDDM achieves state-of-the-art performance across multiple real-world dehazing benchmarks, with significant PSNR and SSIM improvements. Qualitative results demonstrate superior color fidelity and fine-detail preservation, validating its strong cross-domain generalization capability.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Due to the domain gap between real-world and synthetic hazy images, current data-driven dehazing algorithms trained on synthetic datasets perform well on synthetic data but struggle to generalize to real-world scenarios. To address this challenge, we propose extbf{I}mage extbf{D}ehazing extbf{D}iffusion extbf{M}odels (IDDM), a novel diffusion process that incorporates the atmospheric scattering model into noise diffusion. IDDM aims to use the gradual haze formation process to help the denoising Unet robustly learn the distribution of clear images from the conditional input hazy images. We design a specialized training strategy centered around IDDM. Diffusion models are leveraged to bridge the domain gap from synthetic to real-world, while the atmospheric scattering model provides physical guidance for haze formation. During the forward process, IDDM simultaneously introduces haze and noise into clear images, and then robustly separates them during the sampling process. By training with physics-guided information, IDDM shows the ability of domain generalization, and effectively restores the real-world hazy images despite being trained on synthetic datasets. Extensive experiments demonstrate the effectiveness of our method through both quantitative and qualitative comparisons with state-of-the-art approaches.

Problem

• Bridging synthetic-to-real domain gap in image dehazing
• Incorporating physics-guided diffusion for realistic haze removal
• Enhancing generalization of dehazing models to real-world scenarios

Innovation

• Physics-guided diffusion for real-world dehazing
• Incorporates atmospheric scattering into noise diffusion
• Robust domain generalization from synthetic to real