Fixation Guided Integrity Enhancement Network for Salient Object Detection
Yang Shui, Xingwang Wang, Xinze Wang, Yajun Chang, Yuhang Wu, Zelong Wang · 2024
Although significant progress has been made in current salient object detection (SOD), the issue of integrity overflow in complex scene detection remains to be solved. The saliency map can accurately locate salient regions. Therefore, we propose the fixation guided integrity enhancement network (FGIEN). We utilize the fixation guidance module and the integrity enhancement module to achieve accurate localization and salient object integrity learning. The proposed method (FGIEN-F) exhibits better performance on five benchmark datasets. In addition, on the DUTS-F dataset, the proposed method fusing fixation information shows significant improvement in all metrics, with the MAE decreased by 0.015, S-measure, E-measure, and Weighted F-measure increasing by 3.02%, 3.88%, and 6.76%, respectively. This paper verifies the feasibility and superiority of fusing fixation.