ZID-Net: Zero-Inference Diffusion Prior Decoupling Network for Single Image Dehazing

2026-04-26

AI Summary

Single-image dehazing faces a trade-off between restoration quality and computational efficiency: conventional CNNs struggle to model spatially non-uniform haze, while diffusion models suffer from slow inference and unstable sampling. To address this, this work proposes ZID-Net, which decouples diffusion-based supervision from feedforward inference—leveraging a conditional diffusion process during training to provide structural priors, yet relying solely on an efficient feedforward network at test time. The method introduces a novel Zero-Inference Prior Propagation Head (ZI-PPH) to inject diffusion priors into a purely feedforward architecture, alongside a frequency-space decoupled backbone incorporating Channel-Spatial Laplacian Masks (CSLM), a Lightweight Global Context Block (LGCB), and a Dynamic Feature Arbitration Block (DFAB). ZID-Net achieves 40.75 dB PSNR on RESIDE, outperforms state-of-the-art methods by 1.13 dB on real-world data, and improves results by 3.06 dB on StateHaze1k, all with a fast inference time of only 19.35 ms.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Learning to Search
• Machine Learning: Deep Neural Architectures and Foundation Models

Application Category

• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Single image dehazing is often constrained by a trade-off between restoration quality and computational efficiency. While efficient, CNN networks struggle to learn robust priors for dense and non-homogeneous haze. Conversely, diffusion models provide strong generative priors but suffer from severe inference latency and sampling instability. To address these limitations, we propose ZID-Net, a novel framework that explicitly decouples diffusion supervision from feed-forward inference. For efficient inference, we design a frequency-spatial decoupled feed-forward backbone. Within this backbone, a Channel-Spatial Laplacian Mask (CSLM) filters haze-amplified noise to extract purified structural details, while Lightweight Global Context Blocks (LGCBs) establish long-range spatial dependencies to capture the global variations of haze. A Dynamic Feature Arbitration Block (DFAB) then adaptively fuses these semantic and structural features for robust reconstruction. To provide this backbone with physical priors without the inference cost, we introduce a Zero-Inference Prior Propagation Head (ZI-PPH) during training. ZI-PPH leverages a conditional diffusion process to predict residual noise, providing degradation-aware structural supervision to the backbone. By discarding the diffusion branch at test time, ZID-Net integrates diffusion priors into a pure feed-forward architecture for accurate and efficient restoration. ZID-Net achieves 40.75 dB PSNR on the synthetic RESIDE dataset and outperforms existing methods with a 1.13 dB gain on real-world datasets. Additionally, it yields a 3.06 dB PSNR gain on the StateHaze1k remote sensing dataset with an inference time of just 19.35 ms. The project code is available at: https://github.com/XoomitLXH/ZID-Net.

Problem

• single image dehazing
• diffusion models
• computational efficiency
• restoration quality
• inference latency

Innovation

• Zero-Inference Diffusion Prior
• Feed-Forward Dehazing Network
• Frequency-Spatial Decoupling
• Diffusion Supervision Decoupling
• Dynamic Feature Arbitration