Bilateral Feature Fusion with hexagonal attention for robust saliency detection under uncertain environments

Habib Khan, Muhammad Talha Usman, Jakeoung Koo

Abstract: Highlights•The BFF module effectively captures both the global context and local details.•A novel HAF is designed to refine complex features through various attentions.•The cross-feature fusion strategy enhances representations using dual paths.•An uncertainty-aware training ensures the proposed BiFusHNet’s generalization.•Extensive analysis shows the network outperforms SOTA methods on diverse datasets.