Revisiting Weak-to-Strong Consistency in Semi-Supervised Semantic Segmentation

note