Camouflaged Object Detection with CNN-Transformer Harmonization and Calibration

Camouflaged object detection (COD) aims to segment objects that visually blend into their surroundings. However, the subtle differences between camouflaged objects and the background make this task highly challenging. Therefore, how to represent and learn local details and global contexts is crucial for improving detection performance. In this paper, we propose a novel COD network which synergistically leverages the distinct but complementary local and global knowledge to capture the camouflaged objects and identify imperceptible boundaries. Specifically, we design a Feature Coherence Harmonization module to integrate intra-layer features by bridging the knowledge gap between convolutional neural network (CNN) features, which focus on local patterns, and Transformer features, which capture global relationships. Furthermore, we propose a Cross-layer Feature Calibration Module that adaptively aligns inter-layer features, progressively aggregating diverse information to achieve an accurate prediction. Experimental results on COD benchmark datasets demonstrate that the proposed network significantly outperforms state-of-the-art approaches.