Lifelong CycleGAN for continual multi-task image restoration

Yuping Li, Xiangli Nie, Wenhui Diao, Suiwu Zheng

Abstract: Highlights•A novel lifelong CycleGAN (LCGAN) framework is proposed to continually learn multiple image restoration tasks.•Knowledge distillation and memory replay are introduced to alleviate catastrophic forgetting.•LCGAN can be trained with unpaired data and be employed without knowing the task type beforehand.•Experimental results across image enhancement, deblurring and denoising tasks demonstrate the superiority of LCGAN.