Saved in:
Bibliographic Details
Main Authors: Li, Xinheng, Chen, Minghao, Wu, Mengqing, Liu, Yan, Huo, Guanying
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.23709
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908994057011200
author Li, Xinheng
Chen, Minghao
Wu, Mengqing
Liu, Yan
Huo, Guanying
author_facet Li, Xinheng
Chen, Minghao
Wu, Mengqing
Liu, Yan
Huo, Guanying
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2604_23709
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ZID-Net: Zero-Inference Diffusion Prior Decoupling Network for Single Image Dehazing
Li, Xinheng
Chen, Minghao
Wu, Mengqing
Liu, Yan
Huo, Guanying
Computer Vision and Pattern Recognition
Image and Video Processing
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.
title ZID-Net: Zero-Inference Diffusion Prior Decoupling Network for Single Image Dehazing
topic Computer Vision and Pattern Recognition
Image and Video Processing
url https://arxiv.org/abs/2604.23709