Learning Object Bounding Boxes for 3D Instance Segmentation on Point Clouds

TL;DR

This paper introduces 3D-BoNet, a simple, efficient, end-to-end framework for 3D instance segmentation on point clouds that directly predicts bounding boxes and masks without post-processing.

Contribution

The paper presents a novel single-stage, anchor-free approach that directly regresses 3D bounding boxes and masks, significantly improving efficiency and accuracy over prior methods.

Findings

Outperforms existing methods on ScanNet and S3DIS datasets

Achieves approximately 10x computational efficiency

Eliminates need for post-processing steps

Abstract

We propose a novel, conceptually simple and general framework for instance segmentation on 3D point clouds. Our method, called 3D-BoNet, follows the simple design philosophy of per-point multilayer perceptrons (MLPs). The framework directly regresses 3D bounding boxes for all instances in a point cloud, while simultaneously predicting a point-level mask for each instance. It consists of a backbone network followed by two parallel network branches for 1) bounding box regression and 2) point mask prediction. 3D-BoNet is single-stage, anchor-free and end-to-end trainable. Moreover, it is remarkably computationally efficient as, unlike existing approaches, it does not require any post-processing steps such as non-maximum suppression, feature sampling, clustering or voting. Extensive experiments show that our approach surpasses existing work on both ScanNet and S3DIS datasets while being approximately 10x more computationally efficient. Comprehensive ablation studies demonstrate the effectiveness of our design.