Weakly-Supervised Semantic Segmentation Based on Multi-Strategy Collaboration

Fully-supervised semantic segmentation has advanced notably with deep learning, yet its dependence on large amounts of precisely pixel-level annotated data brings high costs, hampering wide application. Weakly-supervised semantic segmentation helps reduce this burden, but current methods still lack in fully covering the target regions and utilizing contextual information. This paper proposes a weakly-supervised semantic segmentation method via multi-strategy collaboration. By combining regional feature expansion, contextual awareness, and feature enhancement fusion, it improves the model’s target recognition and segmentation accuracy. Experiments show that our method surpasses existing techniques on PASCAL VOC 2012 and MS COCO 2014 datasets, offering an effective solution for weakly-supervised semantic segmentation.