CFANet: A Cross-layer Feature Aggregation Network for Camouflaged Object Detection

Deep learning-based camouflaged object detection approaches have achieved great progress in recent years. However, it is still challenging to accurately identify the camouflaged objects from the highly similar backgrounds. In this paper, we design a novel cross-layer feature aggregation network (CFANet) for camouflaged object detection to explore how to effectively aggregate multi-level and multi-scale features generated from the backbone network by excavating the similarities and differences of features at different levels. Firstly, we design a cross-layer feature fusion (CLFF) module to generate discriminative features by fusing and refining the multi-level side-output features with similar characteristics in the same feature group. Secondly, we design a uniqueness enhancement (UE) strategy to respectively emphasize the superiority of deep features and shallow features in locating the camouflaged objects and sharpening the structure details. Extensive experiments on four benchmarks are conducted to demonstrate that the proposed CFANet network performs favorably against 11 state-of-the-art camouflaged object detection methods, demonstrating the effectiveness and superiority of our method. The code and results can be found from the link of https://github.com/ZhangQing0329/CFANet