The IOCI/IOCV dataset is a real-world indoor and outdoor color image (IOCI) and video (IOCV) dataset used for benchmarking. This dataset uses images captured in indoor and outdoor environments using 13 different camera devices. Preset camera settings (such as ISO, shutter speed, and aperture) were not used during acquisition; instead, the cameras' automatic modes were primarily employed. In some uncontrolled low-light environments, the dataset uses short exposure times and increased ISO values ​​to generate noisy images.

The HQ-NightRain dataset includes 10,000 training, 900 validation, and 300 testing image pairs at a resolution of 1280×720. It also provides 512 real nighttime rainy images and 20 synthetic pairs under natural scenes. By incorporating illumination information, HQ-NightRain produces more realistic nighttime rain images covering rain streaks, raindrops, and their mixtures, offering a unified benchmark for developing robust nighttime rain removal methods.

SmokeBench is the first real-world image desmoking dataset focusing on early-stage fire monitoring scenarios. It is constructed using a carefully designed smoke acquisition system to generate a large number of accurately paired clear and smoky images across diverse scenes, providing a valuable foundation for developing and evaluating more robust and practical desmoking algorithms. The release of this dataset is expected to greatly advance the application of AI technologies in fire safety.

The WeatherBench dataset contains 41,402 training and 600 testing image pairs, each with a resolution of 512×512. It features real-world scenes captured under diverse weather conditions (rain, snow, fog) and lighting environments (day and night). WeatherBench serves as a unified benchmark for developing and evaluating all-in-one weather restoration models.

MC-Blur is a large-scale, multi-cause image deblurring dataset that includes both real and synthetic blurred images with various blur factors. The images in this dataset are generated using multiple techniques, including averaging sharp images captured by a high-speed camera at 1000 frames per second, convolving ultra-high-definition (UHD) images with large kernels, applying defocus blur, and collecting real blurred images captured by different camera devices.

The large-scale DPED dataset consists of photos captured simultaneously in outdoor environments using three smartphones and one DSLR camera. The devices used for data collection are the iPhone 3GS, BlackBerry Passport, Sony Xperia Z, and Canon 70D DSLR. To ensure synchronization, all devices were mounted on a tripod and remotely triggered via a wireless control system.

Manga109 is a super-resolution dataset specifically designed for manga images, containing 109 high-quality manga illustrations. This dataset is suitable for studying the super-resolution problem in manga image restoration.

This dataset primarily focuses on plants, buildings, and street scenes, aiming to advance research on high-quality image super-resolution, particularly for outdoor and real-world scenarios.

LWDDS is a synthetic video raindrop dataset. It provides 67,500 image pairs for training from 45 video clips and 600 image pairs for testing from 6 video clips. To enhance diversity, the dataset covers various scenarios and weather conditions, including rural areas, urban streets, highways, and different times of day (morning, noon, and evening), establishing a benchmark for raindrop removal in autonomous driving under adverse weather conditions.