Self-Attention Recurrent Network for Saliency Detection

TL;DR

This paper introduces a novel end-to-end deep saliency network that effectively combines multi-scale feature maps by enhancing local and global information using recurrent structures and self-attention modules, leading to improved saliency detection performance.

Contribution

The paper proposes a new deep saliency network that leverages multi-scale features with recurrent and self-attention modules to better utilize local and global information.

Findings

Outperforms existing algorithms on four datasets

Effectively enhances local features with recurrent structures

Utilizes self-attention for better global context understanding

Abstract

Feature maps in deep neural network generally contain different semantics. Existing methods often omit their characteristics that may lead to sub-optimal results. In this paper, we propose a novel end-to-end deep saliency network which could effectively utilize multi-scale feature maps according to their characteristics. Shallow layers often contain more local information, and deep layers have advantages in global semantics. Therefore, the network generates elaborate saliency maps by enhancing local and global information of feature maps in different layers. On one hand, local information of shallow layers is enhanced by a recurrent structure which shared convolution kernel at different time steps. On the other hand, global information of deep layers is utilized by a self-attention module, which generates different attention weights for salient objects and backgrounds thus achieve better performance. Experimental results on four widely used datasets demonstrate that our method has advantages in performance over existing algorithms.