ReviveDiff: A Universal Diffusion Model for Restoring Images in Adverse Weather Conditions

TL;DR

ReviveDiff introduces a universal diffusion-based framework for restoring images degraded by diverse adverse environments (rain, underwater, low light, smoke, nighttime hazy). It combines a U-shaped latent diffusion model with Coarse-to-Fine Blocks and a Multi-Attentional Feature Complementation module, guided by edge and histogram priors to preserve structure and color. The approach delivers state-of-the-art performance across seven datasets and five degradation types, with strong quantitative gains (PSNR/SSIM/LPIPS) and improved perceptual quality, while maintaining computational efficiency. This work demonstrates the potential of diffusion models for cross-condition image restoration and offers a practical path toward robust vision tasks under challenging conditions.

Abstract

Images captured in challenging environments--such as nighttime, smoke, rainy weather, and underwater--often suffer from significant degradation, resulting in a substantial loss of visual quality. The effective restoration of these degraded images is critical for the subsequent vision tasks. While many existing approaches have successfully incorporated specific priors for individual tasks, these tailored solutions limit their applicability to other degradations. In this work, we propose a universal network architecture, dubbed ``ReviveDiff'', which can address various degradations and bring images back to life by enhancing and restoring their quality. Our approach is inspired by the observation that, unlike degradation caused by movement or electronic issues, quality degradation under adverse conditions primarily stems from natural media (such as fog, water, and low luminance), which generally preserves the original structures of objects. To restore the quality of such images, we leveraged the latest advancements in diffusion models and developed ReviveDiff to restore image quality from both macro and micro levels across some key factors determining image quality, such as sharpness, distortion, noise level, dynamic range, and color accuracy. We rigorously evaluated ReviveDiff on seven benchmark datasets covering five types of degrading conditions: Rainy, Underwater, Low-light, Smoke, and Nighttime Hazy. Our experimental results demonstrate that ReviveDiff outperforms the state-of-the-art methods both quantitatively and visually.

Figures (9)

• Figure 1: Our ReviveDiff can address various adverse degradations (such as rain, low light, underwater, and nighttime dehazing) by enhancing and restoring their quality.
• Figure 2: The basic framework of our proposed ReviveDiff, which is a U-shaped latent diffusion model with stacked Coarse-to-Fine Learning Blocks (C2FBlock) and Multi-Attentional Feature Complementation (MAFC) modules, guided by a combined loss to tackle diverse real-world image restoration tasks.
• Figure 3: The Coarse-to-Fine Learning Block (C2FBlock) at the core of our proposed ReviveDiff network.
• Figure 4: Our Multi-Attentional Feature Complementation (MAFC) module adaptively integrates coarse and fine features through complementary refinement pathways. The coarse stream undergoes dual-domain enhancement via channel-wise recalibration and spatial context aggregation, while a point-wise feature refinement module refines the fine branch. The two branches are then fused through an adaptive weighting strategy, resulting in complemented feature maps that preserve both global context and local details.
• Figure 5: Visual comparison of the deraining results obtained with our ReviveDiff and SOTA approaches on Rain100L Rain100. The small images enclosed in red below each result provide an enlarged view of the areas highlighted by red rectangles.