Leveraging a Cross-Supervision SAM for Weakly-Supervised Camouflaged Object Detection

Camouflaged Object Detection (COD) has been attracting great attention as a result of its significant applications in various fields. Limited by the high cost of annotation, scribble annotation as a weakly-supervised approach is a promising solution. Previous methods focused mainly on detecting camouflaged objects using local features. However, such a solution tends to produce inaccurate detection when dealing with highly camouflaged objects because of the neglect of global context. In this paper, we propose a dual-branch cross-supervision framework based on the Segment Anything Model (SAM). The architecture includes two branches: one relies solely on global features for learning, while the other uses scribble annotations as prompts to guide training with local semantic information. By integrating the two branches, the model can more accurately capture the location and boundary information of camouflaged objects. The extensive experimental results demonstrate that our method achieves new state-of-the-art performances for weakly-supervised COD methods, significantly improving different baselines. The code will be released.