Camouflaged Object Detection with Discriminative Information Attention and Cross-level Feature Fusion

Camouflaged object detection (COD) aims to accurately locate and segment objects that blend in with surroundings, which is a challenging task due to the inconspicuous appearance and boundary. In this paper, we propose a novel Discriminative Information Attention and Cross-level Feature Fusion Network (DACF-Net) for camouflaged object detection. Specifically, we introduce a Discriminative Information Attention Module (DIAM) and a Context Enriched Module(CEM) to exploit and extract representative and contextual information. A Cross-level Feature Aggregation Module (CFAM) is utilized to suppress feature redundancies and exploit more complementary information. Extensive quantitative and qualitative experiments indicate that the proposed method has superior performance in explaining the camouflage prediction.