Fusion Boundary and Gradient Enhancement Network for Camouflage Object Detection

The problems of boundary interruption and missing internal texture feature have not been well solved in the current camouflaged object detection model, and the parameters of the model are generally large. To overcome these challenges, we propose a fusion boundary and gradient enhancement networks BGENet, which guides the context features by gradient features and boundary features together. BGENet is divided into three branches, context feature branch, boundary feature branch and gradient feature branch. Furthermore, a parallel context information enhancement module is introduced to enhance the context features. The designed pre-background information interaction module is used to highlight the boundary features of the camouflaged object and guide the context features to compensate for the boundary breaks in the context features, while we use the learned gradient features to guide the context features through the proposed gradient guidance module, and enhances internal information about context features. Experiments on CAMO, COD10K and NC4K three datasets confirm the effectiveness of our BGENet, which uses only 20.81M parameters and achieves superior performance compared with traditional and SOTA methods.