Referring flexible image restoration

Runwei Guan, Rongsheng Hu, Zhuhao Zhou, Tianlang Xue, Ka Lok Man, Jeremy S. Smith, Eng Gee Lim, Weiping Ding, Yutao Yue

Abstract: Highlights•A novel task: Referring Flexible Image Restoration, prompt-based degrade-removal.•A novel dataset called RFIR, containing 153,423 samples with five degradation types.•A multi-task model for specific degradation-removal and multi-deg classification.•A highly efficient attention module called MHASA with linear complexity.•A novel fusion method called MHACA with better performance than vanilla MHCA.