Weakly supervised semantic segmentation with multi-task learning and segment anything model

doi:10.1016/j.neucom.2025.131131

Weakly supervised semantic segmentation with multi-task learning and segment anything model

2025-07-29

0

PRE

AI

DOI：10.1016/j.neucom.2025.131131

Original

Original request for help

Save

Abstract

En 中文

• Novel CAMs generation network enhances localization precision through multi-task learning. • SAM-guided training strategy reduces label noise sensitivity by filtering pseudo-labels. • Comprehensive ablation studies validate the effectiveness of components. • Integration of SAM’s boundary knowledge into WSSS training pipeline.

Keywords:

CAMs

multi-task learning

SAM

WSSS

ablation study

Weakly supervised semantic segmentation with multi-task learning and segment anything model

Abstract

Journal

Organization