CORNet: A Consistency-based Outlier Rejection Network for non-rigid registration

Abstract

Non-rigid point cloud registration is an important problem in computer vision and graphics, aiming to find the warping function between deformed point clouds. In this paper, we propose CORNet, a consistency-based outlier rejection network for non-rigid registration. By leveraging the local geometric structure and probability distribution of point clouds, we obtain local spatial consistency and Gaussian probabilistic consistency. We then employ the Transformer mechanism, combined with consistency information, to classify inliers and outliers in correspondences, ultimately obtaining high-quality correspondences for non-rigid registration. Ablation studies validate the effectiveness of our method, and extensive experiments demonstrate that our method achieves state-of-the-art performance.