EPANet: Edge-assisted Position Aware Attention Network for Camouflaged Object Detection

Camouflaged object detection (COD), where the object blends in with its surroundings, makes it a challenging task. The features extracted by Res2Net-50 are excellent in terms of detail, but slightly lacking for the extraction of semantic information. So we propose an position aware attention network. We design a position aware attention module in order to model the correlation between high-level features and between pixels. This module can effectively address the shortcomings of Res2Net. Also, we propose a semantic guidance feature cascade module. Guided by the top-level features, the refined features can be effectively fused layer by layer. We demonstrate the superiority of our proposed method over the other 8 state-of-the-art methods on three datasets.