LL-Bench: Rethinking Low-Level Vision Evaluation in the Era of Large-Scale Generative Models | ArxivCSExplorer