WeatherBench is a dataset released in 2025 by Dalian University of Technology in collaboration with Nanjing University of Science and Technology and Dalian Maritime University. It is designed for image restoration tasks under real-world adverse weather conditions. The related paper is titled WeatherBench: A Real-World Benchmark Dataset for All-in-One Adverse Weather Image Restoration. It aims to provide a unified, realistic and large-scale training and evaluation benchmark for all-in-one image restoration models such as rain removal, snow removal, and fog removal.

This dataset contains 50,000 pairs of degraded images from severe weather and their corresponding clear images. After quality screening, 42,002 high-quality paired samples were retained, of which 41,402 pairs were used for training and 600 pairs were used for testing. All images were uniformly cropped to a resolution of 512 × 512 to facilitate model training and fair comparison.

Data composition:

• Sample format: Paired data of strictly aligned degraded images (LQ) and sharp reference images (GT)
• Weather degradation types: Rain, Snow, and Haze.
• Lighting conditions: Daytime and nighttime scenes