Abstract: Due to the high similarity between the camouflage objects and their surrounding environment in colors, textures, or patterns, it is quite difficult for the networks to identify the indistinguishable boundaries. Recent camouflaged object segmentation models usually directly introduce boundary detection task as supplement to enhance the detection ability of subtle structures and details. However, such predicted boundaries are often inaccurate or even incorrect. It is not an optimal solution to directly integrate the predicted boundaries at early stage with the object features at different levels for precise segmentation. To address this problem, we propose a bidirectional edge object interaction network, which consists of the bidirectional spatial attention guidance module, the bidirectional feature complementation module, and feature collaborative calibration module. Our proposed network explores the complementary information between the boundary features and the object features, and gradually corrects the errors in an interactive learning manner, thereby achieving more accurate predictions. We conduct extensive experiments on four widely used benchmarks and compare with several state-of-the-arts methods. The experimental results demonstrate the effectiveness and superiority of our network.