Weakly supervised semantic segmentation (WSSS) seeks to assign a class label to every pixel in an image while relying on annotations that are coarser, cheaper or easier to obtain than full pixel-level ...