3D Correspondence Grouping with Compatibility Features

TL;DR

This paper introduces a novel compatibility feature (CF) representation for 3D correspondence grouping, leveraging geometric constraints and a simple neural network to improve inlier classification accuracy.

Contribution

It proposes a new CF representation for 3D correspondences and formulates the grouping as a classification task solved by an MLP, outperforming existing methods.

Findings

CF is robust and rotation-invariant

The method achieves state-of-the-art performance

Good generalization across benchmarks

Abstract

We present a simple yet effective method for 3D correspondence grouping. The objective is to accurately classify initial correspondences obtained by matching local geometric descriptors into inliers and outliers. Although the spatial distribution of correspondences is irregular, inliers are expected to be geometrically compatible with each other. Based on such observation, we propose a novel representation for 3D correspondences, dubbed compatibility feature (CF), to describe the consistencies within inliers and inconsistencies within outliers. CF consists of top-ranked compatibility scores of a candidate to other correspondences, which purely relies on robust and rotation-invariant geometric constraints. We then formulate the grouping problem as a classification problem for CF features, which is accomplished via a simple multilayer perceptron (MLP) network. Comparisons with nine state-of-the-art methods on four benchmarks demonstrate that: 1) CF is distinctive, robust, and rotation-invariant; 2) our CF-based method achieves the best overall performance and holds good generalization ability.