Abstract: Class-incremental semantic segmentation focuses on updating the segmentation model with only new-class samples. Catastrophic forgetting and background shift are the two prevalent challenges.