Figure 1. The framework of Class-balanced Pixel-level Self-Labeling (CPSL). The model contains a main segmentation network f and
SEG
its momentum-updated version f′ . The f′ is followed by a self-labeling head f and its momentum version f′ , which projects
SEG SEG SL SL
pixel-wise feature embedding into a class probability vector. The pixel-level self-labeling module produces soft cluster assignment P to
SL
gradually rectify soft pseudo label P . Then the segmentation loss Lt is computed between the prediction map P and the rectified
ST SEG
pseudo label Yˆt. To train the self-labeling head, we randomly sample pixels from each image, and use the memory bank M, which
contains previous batches of pixel features, to augment the current batch. Then we compute the optimal transport assignment Q over
aug
the augmented data by enforcing class balance, and use the assignment of current batch Q to compute the self-labeling loss L .
cur SL