Weakly supervised instance segmentation using multi-stage erasing refinement and saliency-guided proposals ordering

Zheng Hu, Zhi Liu, Gongyang Li, Linwei Ye, Lei Zhou, Yang Wang

Abstract: Highlights•We propose a novel weakly supervised instance segmentation framework equipped with MSER and SGPO.•We propose the Multi-Stage Erasing Refinement method to better provide location and boundary information of objects.•We propose the Saliency-Guided Proposals Ordering method to alleviate the adverse effect of background regions.•Experiments on PASCAL VOC 2012 and COCO datasets demonstrate the superiority of our method.