PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation

  • 2020-04-03 16:26:37
  • Li Jiang, Hengshuang Zhao, Shaoshuai Shi, Shu Liu, Chi-Wing Fu, Jiaya Jia
  • 4

Abstract

Instance segmentation is an important task for scene understanding. Comparedto the fully-developed 2D, 3D instance segmentation for point clouds have muchroom to improve. In this paper, we present PointGroup, a new end-to-endbottom-up architecture, specifically focused on better grouping the points byexploring the void space between objects. We design a two-branch network toextract point features and predict semantic labels and offsets, for shiftingeach point towards its respective instance centroid. A clustering component isfollowed to utilize both the original and offset-shifted point coordinate sets,taking advantage of their complementary strength. Further, we formulate theScoreNet to evaluate the candidate instances, followed by the Non-MaximumSuppression (NMS) to remove duplicates. We conduct extensive experiments on twochallenging datasets, ScanNet v2 and S3DIS, on which our method achieves thehighest performance, 63.6% and 64.0%, compared to 54.9% and 54.4% achieved byformer best solutions in terms of mAP with IoU threshold 0.5.

 

Quick Read (beta)

loading the full paper ...