Simultaneously Localize, Segment and Rank the Camouflaged Objects

Camouflage is a key defence mechanism across species that is critical tosurvival. Common strategies for camouflage include background matching,imitating the color and pattern of the environment, and disruptive coloration,disguising body outlines [35]. Camouflaged object detection (COD) aims tosegment camouflaged objects hiding in their surroundings. Existing COD modelsare built upon binary ground truth to segment the camouflaged objects withoutillustrating the level of camouflage. In this paper, we revisit this task andargue that explicitly modeling the conspicuousness of camouflaged objectsagainst their particular backgrounds can not only lead to a betterunderstanding about camouflage and evolution of animals, but also provideguidance to design more sophisticated camouflage techniques. Furthermore, weobserve that it is some specific parts of the camouflaged objects that makethem detectable by predators. With the above understanding about camouflagedobjects, we present the first ranking based COD network (Rank-Net) tosimultaneously localize, segment and rank camouflaged objects. The localizationmodel is proposed to find the discriminative regions that make the camouflagedobject obvious. The segmentation model segments the full scope of thecamouflaged objects. And, the ranking model infers the detectability ofdifferent camouflaged objects. Moreover, we contribute a large COD testing setto evaluate the generalization ability of COD models. Experimental results showthat our model achieves new state-of-the-art, leading to a more interpretableCOD network.