RML: Efficient Representation Mutual Learning Framework for End-to-End Weakly Supervised Semantic Segmentation

Rongtao Xu, Changwei Wang, Shibiao Xu, Weiliang Meng, Xiaopeng Zhang

Abstract: Research on efficient semantic segmentation models is increasing the number of instrumentation and measurement applications. In recent years, there has been significant progress in the development of end-to-end solutions for weakly supervised semantic segmentation (WSSS) using image-level labels. Previous end-to-end methods lack adequate interaction of multilevel representation information and typically rely on a segmentation branch and a pseudolabel segmentation loss to predict segmentation masks, leading to increased time consumption and parameter count. In this article, we propose an effective representation mutual learning (RML) framework to directly predict segmentation masks. This framework leverages collaborative learning and mutual teaching among multilevel features to improve segmentation performance. Our RML incorporates the category-level, feature-level, and pixel-level RML strategies to improve the segmentation quality. To improve the class activation map (CAM) representations, we propose a CAM-driven category-level mutual learning (CCML) strategy. In addition, we design a multiscale feature-level mutual learning (MFML) strategy, which can align aggregated contextual feature representations. Furthermore, we also provide an affinity-aware pixel-level mutual learning (APML) strategy to learn semantic affinity representations. Our experiments show that the RML framework outperforms state-of-the-art methods in terms of time efficiency and accuracy.

External IDs:dblp:journals/tim/XuWXMZ25