Camouflaged Object Detection via Global-Edge Context and Mixed-Scale Refinement

Camouflage object detection (COD), trying to segment objects that blend perfectly with the surrounding environment, is challenging and complex in real-world scenarios. However, existing deep learning methods often fail to accurately identify the boundary detail branches and complete structure of camouflaged objects. To address these challenges, we propose a novel Global-edge Context and Mixed-scale Refinement Network (GCMRNet) to handle the challenging COD task. Specifically, we propose a Global-edge Context Module (GCM), to effectively obtain long-distance context dependencies, which provides rich global context information. In addition, we also propose the Hierarchical Mixed-scale Refinement Module (HMRM) to conduct information interaction and feature refinement between channels, which aggregates the rich multi-level features for accurate COD. Extensive experimental results on three challenging benchmark datasets demonstrate that our proposed method outperforms 21 state-of-art methods under widely used evaluation metrics.