Closed shengyi4 closed 2 years ago
I think it can't support because MMSelfSup is designed for upstream self-supervised learning and semi-supervised semantic segmentation usually focuses on partial images with labels and the other without labels, where most algorithms are proposed on relation between fully supervised part and self supervised part.
Can this approach be used in semi-supervised image segmentation, such as deeplabv3, segformer, and so on? Thanks!