Deep Models with Fusion Strategies for MVP Point Cloud Registration

TL;DR

This paper presents a fusion-based deep learning approach combining ROPNet and PREDATOR models to improve point cloud registration in challenging MVP scenarios, achieving high accuracy and robustness.

Contribution

We propose a novel ensemble strategy that fuses two deep models for MVP point cloud registration, addressing challenges like low overlap and non-uniform density.

Findings

Achieved second place in MVP registration challenge

Attained low error metrics: Rot_Error 2.96546, Trans_Error 0.02632, MSE 0.07808

Demonstrated effectiveness of model fusion in complex registration tasks

Abstract

The main goal of point cloud registration in Multi-View Partial (MVP) Challenge 2021 is to estimate a rigid transformation to align a point cloud pair. The pairs in this competition have the characteristics of low overlap, non-uniform density, unrestricted rotations and ambiguity, which pose a huge challenge to the registration task. In this report, we introduce our solution to the registration task, which fuses two deep learning models: ROPNet and PREDATOR, with customized ensemble strategies. Finally, we achieved the second place in the registration track with 2.96546, 0.02632 and 0.07808 under the the metrics of Rot\_Error, Trans\_Error and MSE, respectively.