Code and trained weights for S²UG-Mamba

Aiming to address the challenges in visual saliency prediction, including high computational cost for long-range modeling, over-propagated responses to complex backgrounds, and dispersed fixation hotspots caused by decoder upsampling, we propose a novel saliency prediction network named Spatio-Spectral Uncertainty-Gated Mamba Network (S²UG-Mamba).

This model integrates uncertainty-gated Mamba enhancement and dynamic frequency domain modulation.