KwaiSR Dataset
收藏资源简介:
KwaiSR数据集是中国科学院大学和快手技术公司合作创建的第一个针对野外短格式UGC图像超分辨率任务的基准数据集。该数据集包含合成和野外两部分,合成部分由模拟真实世界低质量短格式UGC图像退化的分布产生,共有1900对图像;野外部分则是直接从快手平台收集的低质量图像,经过质量评估方法KVQ筛选得到1900张图像。数据集分为训练、验证和测试集,比例为8:1:1。KwaiSR数据集用于NTIRE 2025挑战赛,旨在通过扩散型/生成型方法提高低质量短格式UGC图像的主观质量。
The KwaiSR dataset is the first benchmark dataset for wild short-form UGC (User-Generated Content) image super-resolution tasks, collaboratively developed by the University of Chinese Academy of Sciences (UCAS) and Kuaishou Technology. This dataset comprises two subsets: the synthetic subset and the wild subset. The synthetic subset is generated by simulating the degradation distribution of real-world low-quality short-form UGC images, containing a total of 1900 image pairs. The wild subset consists of low-quality images directly collected from the Kuaishou platform, with 1900 images selected via the quality assessment method KVQ. The dataset is split into training, validation and test sets at a ratio of 8:1:1. The KwaiSR dataset is utilized for the NTIRE 2025 Challenge, aiming to improve the subjective quality of low-quality short-form UGC images through diffusion-based or generative methods.




