Salient object detection based on multi-layer progressive fusion

Abstract

Abstract How to integrate the features of different layers plays an important role in current research of salient object detection. In order to inherit the useful features of various layers, the multi-layer progressive fusion(MLPF) model is proposed in this paper. Specifically, the model first modified the multi-scale enrichment module(MSEM) to enrich the features of adjacent layers. Then, the adjacent feature aggregation module(AFAM) is proposed to complement the feature of adjacent layers. At the same time, to make full use of the features of various layers, the cross-layer feature aggregation module(CFAM) is designed to retain the detail and semantic features and improve the representation for the salient object. Extensive experiments on four public datasets demonstrate that the proposed model outperforms the other state-of-the-art methods.