MZUAI和IIAI在NIPS2023上的一篇论文,用prompt来编码degradation,然后用来guide restoration network,使得模型能够泛化到不同degradation types and levels,也就是说是一个模型一次训练能够应对多种degradation的unified model。文章分析,对每种degradation都要train一个模型是compute-intensive and tedious process, and oftentimes impra