Zevin2023/SEAGULL-100w

SEAGULL-100w是一个大规模合成的基于感兴趣区域（ROI）的图像质量评估（IQA）数据集，用于“SEAGULL：通过视觉-语言指令调优的无参考图像质量评估”。该数据集包含六种不同类型的失真图像，分别是模糊、锐度、曝光、对比度、色彩丰富度和压缩。每种失真类型有二十个不同的参数，总共生成978,720张失真图像。数据集结构包括失真图像、参考图像和标签，标签中提供了ROI的语义描述和对象语义。SEAGULL-100w is a large-scale synthetic ROI-based no-reference image quality assessment dataset, containing six distortion types and twenty parameters, generating a total of 978,720 distorted images. The dataset structure includes distorted images, reference images, and label files, which provide semantic descriptions and object semantics for ROIs.