Yan Jinnan ; Le Trung-Nghia ; Nguyen Khanh-Duy ; Tran Minh-Triet ; Do Thanh-Toan ; Nguyen Tam V.

Abstract

Camouflaged objects are generally difficult to be detected in their naturalenvironment even for human beings. In this paper, we propose a novelbio-inspired network, named the MirrorNet, that leverages both instancesegmentation and mirror stream for the camouflaged object segmentation.Differently from existing networks for segmentation, our proposed networkpossesses two segmentation streams: the main stream and the mirror streamcorresponding with the original image and its flipped image, respectively. Theoutput from the mirror stream is then fused into the main stream's result forthe final camouflage map to boost up the segmentation accuracy. Extensiveexperiments conducted on the public CAMO dataset demonstrate the effectivenessof our proposed network. Our proposed method achieves 89% in accuracy,outperforming the state-of-the-arts. Project Page: https://sites.google.com/view/ltnghia/research/camo