Attention-guided Unified Network for Panoptic Segmentation

  • 2018-12-10 16:25:10
  • Yanwei Li, Xinze Chen, Zheng Zhu, Lingxi Xie, Guan Huang, Dalong Du, Xingang Wang
  • 4

Abstract

This paper studies panoptic segmentation, a recently proposed task whichsegments foreground (FG) objects at the instance level as well as background(BG) contents at the semantic level. Existing methods mostly dealt with thesetwo problems separately, but in this paper, we reveal the underlyingrelationship between them, in particular, FG objects provide complementary cuesto assist BG understanding. Our approach, named the Attention-guided UnifiedNetwork (AUNet), is an unified framework with two branches for FG and BGsegmentation simultaneously. Two sources of attentions are added to the BGbranch, namely, RPN and FG segmentation mask to provide object-level andpixel-level attentions, respectively. Our approach is generalized to differentbackbones with consistent accuracy gain in both FG and BG segmentation, andalso sets new state-of-the-arts in the MS-COCO (46.5% PQ) benchmarks.

 

Quick Read (beta)

loading the full paper ...