Cross Stage Interweaved Fusion Network for Salient Object Detection
Jiachen Yu, Lanfang Miao, Liyuan Chen, Zheng Zheng · 2021 IEEE 3rd International Conference on Frontiers Technology of Information and Computer (ICFTIC) · 2021
Salient object detection, in the artificial intelligence field, is a fundamental and significant computer vision (CV) task which intends to recognize and segment the most remarkable objects. This paper proposes a novel SOD architecture named Cross Stage Interweaved Fusion Network (CSIFNet) which considers features both in the first stage of encoder and the second stage of decoder. In a specific, we employed the following four modules: Global Feature Interweave (GFI) module, Attentional Guidance Head (AGH) module, Residual Aggregation (RA) module, and Self Enhance (SE) module to construct the architecture of cross-stage and multi-scale feature aggregation. By adopting the designed strategy, the detailed features can be indirectly passed through to deeper layers between two main stages. Compared with nine state-of-the-art salient object detectors, CSIFNet performs well on five public datasets.