Panoptic SegFormer

  • 2021-09-08 17:59:12
  • Zhiqi Li, Wenhai Wang, Enze Xie, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Tong Lu, Ping Luo
  • 38

Abstract

We present Panoptic SegFormer, a general framework for end-to-end panopticsegmentation with Transformers. The proposed method extends Deformable DETRwith a unified mask prediction workflow for both things and stuff, making thepanoptic segmentation pipeline concise and effective. With a ResNet-50backbone, our method achieves 50.0\% PQ on the COCO test-dev split, surpassingprevious state-of-the-art methods by significant margins without bells andwhistles. Using a more powerful PVTv2-B5 backbone, Panoptic-SegFormer achievesa new record of 54.1\%PQ and 54.4\% PQ on the COCO val and test-dev splits withsingle scale input.

 

Quick Read (beta)

loading the full paper ...