Yu Qian ; Zhao Xiaoqi ; Pang Youwei ; Zhang Lihe ; Lu Huchuan

Abstract

Dichotomous Image Segmentation (DIS) has recently emerged towardshigh-precision object segmentation from high-resolution natural images. When designing an effective DIS model, the main challenge is how to balancethe semantic dispersion of high-resolution targets in the small receptive fieldand the loss of high-precision details in the large receptive field. Existingmethods rely on tedious multiple encoder-decoder streams and stages togradually complete the global localization and local refinement. Human visual system captures regions of interest by observing them frommultiple views. Inspired by it, we model DIS as a multi-view object perceptionproblem and provide a parsimonious multi-view aggregation network (MVANet),which unifies the feature fusion of the distant view and close-up view into asingle stream with one encoder-decoder structure. With the help of the proposedmulti-view complementary localization and refinement modules, our approachestablished long-range, profound visual interactions across multiple views,allowing the features of the detailed close-up view to focus on highly slenderstructures.Experiments on the popular DIS-5K dataset show that our MVANetsignificantly outperforms state-of-the-art methods in both accuracy and speed.The source code and datasets will be publicly available at\href{https://github.com/qianyu-dlut/MVANet}{MVANet}.