Multi-Modal Salient Feature Enhance for Rgb-T Salient Object Detection

Abstract

Semantic information is essential in RGB-T salient object detection (SOD). Most existing methods directly input the extracted low-level features into the interaction module and utilize a simple recursive structure for high-level semantic guidance. Despite their excellent performance in several scenarios, they suffer from capturing and exploiting the attributes and complementary potential between different feature layers of images, which are critical in obtaining greater details and accurate object location. In this work, we proposed a network for better detail preservation and accurate object location in SOD. On the one hand, a Salient Features Enhanced (SFE) constituted by a multi-branch structure is presented to serve as a bridge between encoding and cross-modality decoding to improve the object details representation. On the other hand, a High-level Semantic Guide (HSGB) constituted by channel attention and a multi-branch structure is designed to guide the cross-modality interaction module and retain the object location information. Evaluation results on three common benchmark datasets reveal that our method achieves competitive state-of-the-art performance.

Keywords: Salient objects detection, SFE, HSGB, Details information, Location information