Towards Fine-grained Large Object Segmentation 1st Place Solution to 3D AI Challenge 2020 -- Instance Segmentation Track

TL;DR

This paper presents a top-performing 3D instance segmentation solution using an ensemble of PointRend models, effectively handling large objects in 3D point cloud data, and achieving first place in the 3D AI Challenge 2020.

Contribution

It introduces a novel ensemble approach with PointRend for fine-grained large object segmentation in 3D point clouds, outperforming existing methods.

Findings

Achieved 1st place in 3D AI Challenge 2020

Ensemble of 5 PointRend models improves segmentation quality

Outperforms HTC and SOLOv2 on large object segmentation

Abstract

This technical report introduces our solutions of Team 'FineGrainedSeg' for Instance Segmentation track in 3D AI Challenge 2020. In order to handle extremely large objects in 3D-FUTURE, we adopt PointRend as our basic framework, which outputs more fine-grained masks compared to HTC and SOLOv2. Our final submission is an ensemble of 5 PointRend models, which achieves the 1st place on both validation and test leaderboards. The code is available at https://github.com/zehuichen123/3DFuture_ins_seg.