Hybrid Attention Asynchronous Cascade Network for Salient Object Detection

The highlighted area or object is defined as the salient region object. For detection, main challenges are still clarity of boundary information and positioning accuracy in complex background, such noise occlusion. To remedy these issues, it proposed that asynchronous cascade saliency detection algorithm based on a deep network, which embedded an encoder–decoder architecture. Moreover, lightweight hybrid attention module designed to obtain explicit boundaries regions. In order effectively improve location objects, this paper adopts bi-directional fusion strategy, generates prediction maps with higher accuracy. experimental results five benchmark datasets show network HACNet par state art for image datasets.