WeatherBench: A Real-World Benchmark Dataset for All-in-One Adverse Weather Image Restoration

2025-09-15

AI Summary

Existing methods rely on synthetic single-weather datasets, suffering from significant domain shift and insufficient weather diversity; critically, the absence of authentic multi-weather image pairs severely hinders unified model development and fair evaluation. To address this, we introduce RealWeather—the first real-scenario, all-weather integrated image restoration benchmark—comprising high-quality, pixel-aligned degraded–clean image pairs under rain, snow, fog, and other adverse conditions. Leveraging multi-condition field photography and sub-pixel-level alignment, RealWeather ensures data authenticity and geometric consistency. It supports both supervised learning and cross-weather generalization studies. The dataset is publicly released and employed in systematic benchmarking, exposing critical performance bottlenecks of state-of-the-art methods under realistic, complex weather conditions. RealWeather establishes a reproducible, extensible foundation for robust image restoration research.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Multimodal Learning

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• 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
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Existing all-in-one image restoration approaches, which aim to handle multiple weather degradations within a single framework, are predominantly trained and evaluated using mixed single-weather synthetic datasets. However, these datasets often differ significantly in resolution, style, and domain characteristics, leading to substantial domain gaps that hinder the development and fair evaluation of unified models. Furthermore, the lack of a large-scale, real-world all-in-one weather restoration dataset remains a critical bottleneck in advancing this field. To address these limitations, we present a real-world all-in-one adverse weather image restoration benchmark dataset, which contains image pairs captured under various weather conditions, including rain, snow, and haze, as well as diverse outdoor scenes and illumination settings. The resulting dataset provides precisely aligned degraded and clean images, enabling supervised learning and rigorous evaluation. We conduct comprehensive experiments by benchmarking a variety of task-specific, task-general, and all-in-one restoration methods on our dataset. Our dataset offers a valuable foundation for advancing robust and practical all-in-one image restoration in real-world scenarios. The dataset has been publicly released and is available at https://github.com/guanqiyuan/WeatherBench.

Problem

• Addressing domain gaps in all-in-one weather image restoration
• Providing real-world benchmark dataset for multiple weather degradations
• Enabling supervised learning with aligned degraded-clean image pairs

Innovation

• Real-world benchmark dataset development
• Supervised learning with aligned image pairs
• Comprehensive benchmarking of restoration methods