Region-based online selective examination for weakly supervised semantic segmentation

Qi Chen, Yun Chen, Yuheng Huang, Xiaohua Xie, Lingxiao Yang

Abstract: Highlights•An online noise correction method for training segmentation models with noisy pseudo-labels.•A SAM-based region voting strategy for pixel-to-region aggregation.•A cross-view examination for adaptively correcting noisy pseudo-labels.•Our method achieves state-of-the-art performance on main benchmarks.