Pang Youwei ; Zhao Xiaoqi ; Xiang Tian-Zhu ; Zhang Lihe ; Lu Huchuan

Abstract

Recent camouflaged object detection (COD) attempts to segment objectsvisually blended into their surroundings, which is extremely complex anddifficult in real-world scenarios. Apart from the high intrinsic similaritybetween camouflaged objects and their background, objects are usually diversein scale, fuzzy in appearance, and even severely occluded. To this end, wepropose an effective unified collaborative pyramid network that mimics humanbehavior when observing vague images and videos, \ie zooming in and out.Specifically, our approach employs the zooming strategy to learn discriminativemixed-scale semantics by the multi-head scale integration and rich granularityperception units, which are designed to fully explore imperceptible cluesbetween candidate objects and background surroundings. The former's intrinsicmulti-head aggregation provides more diverse visual patterns. The latter'srouting mechanism can effectively propagate inter-frame differences inspatiotemporal scenarios and be adaptively deactivated and output all-zeroresults for static representations. They provide a solid foundation forrealizing a unified architecture for static and dynamic COD. Moreover,considering the uncertainty and ambiguity derived from indistinguishabletextures, we construct a simple yet effective regularization, uncertaintyawareness loss, to encourage predictions with higher confidence in candidateregions. Our highly task-friendly framework consistently outperforms existingstate-of-the-art methods in image and video COD benchmarks. Our code can befound at {https://github.com/lartpang/ZoomNeXt}.