Scale-adaptive multi-predator network for camouflaged object detection

In this paper, we propose a scale-adaptive multi-predator (SAMP) network for camouflaged object detection (COD), which aims to segment objects that have successfully concealed themselves by blending into the surrounding environment. Despite the challenging nature of distinguishing objects from a highly similar background, most existing works solve COD as a pixel-level binary classification problem and use a single classifier to identify the target object, analogous to deploying one predator to prey the target animal. Inspired by the high effectiveness of group predation (e.g., wolves), we propose to simultaneously employ a group of diverse “predators” to seek out the target object. To this end, we reformulate COD as a point-to-set matching problem and classify each pixel based on its distance with a set of dynamic query embeddings. Furthermore, we employ a confidence-based query selection method to enable each query to adaptively focus on different scales of the image for better identification of the camouflaged object. Extensive experiments on the CHAMELEON, CAMO, and COD10K datasets verify the effectiveness of the proposed method.