Deep Models with Fusion Strategies for MVP Point Cloud Registration
This is an incremental improvement for point cloud registration in computer vision, addressing low overlap and non-uniform density in a specific competition.
The paper tackled the Multi-View Partial point cloud registration challenge by fusing ROPNet and PREDATOR deep learning models with custom ensemble strategies, achieving second place with metrics of 2.96546 Rot_Error, 0.02632 Trans_Error, and 0.07808 MSE.
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.