Weighted Dense Semantic Aggregation and Explicit Boundary Modeling for Camouflaged Object Detection
Weiyun Liang, Jiesheng Wu, Xinyue Mu, Fangwei Hao, Ji Du, Jing Xu, Ping Li
Abstract
Camouflaged object detection in monocular images has garnered broad attention recently, aiming to segment objects that have high intrinsic similarity with their surroundings. Despite remarkable performance achieved by existing methods, two limitations persist: insufficient utilization of multi-level semantics at each decoding scale and a lack of “explicit” knowledge guidance in boundary learning, leading to performance drops in challenging scenarios. To address these issues, we propose a weighted dense semantic aggregation and explicit boundary modeling network. Specifically, a weighted dense semantic aggregation module is proposed to sufficiently aggregate multi-level semantics at each decoding scale, and enable the exploration of the relationship between multi-level features and camouflaged objects. An explicit boundary modeling module is developed to capture edge semantics with explicit boundary knowledge guidance and enhance the feature representation with edge cues. A detail enhanced multi-scale module is further designed to refine multi-scale features. Extensive experiments demonstrate that our proposed method achieves competitive performance against state-of-the-art methods on four benchmark datasets without excessive model complexity. Codes and results will be released at https://github.com/crrcoo/SAE-Net.