Co-Saliency Detection Based on Hierarchical Interactive Attention and Feedback Enhancement

Co-saliency object detection (CoSOD) aims to detect common salient objects in a set of related images. Existing methods generally focus on mining consistency or correlation information between image groups, but many methods ignore the influence of non-salient objects or backgrounds, which may affect the model's ability to distinguish irrelevant objects. We propose a co-saliency detection algorithm based on hierarchical interactive attention and feedback enhancement. It integrates token and self-attention interaction to automatically aggregate the consistency between images, and utilizes global consistency feature feedback to enhance the feature representation of images by combining the interaction between channel and spatial attention. The quantitative comparison and qualitative analysis on three benchmark datasets indicate that our algorithm has achieved significant improvements in key indicators such as E-measure, F-measure, and S-measure.