Feature Query Networks: Neural Surface Description for Camera Pose Refinement
Hugo Germain, Daniel DeTone, Geoffrey Pascoe, Tanner Schmidt, David Novotný, Richard Newcombe, Chris Sweeney, Richard Szeliski, Vasileios Balntas
Abstract
Accurate 6-DoF camera pose estimation in known environments can be a very challenging task, especially when the query image was captured at viewpoints strongly differing from the set of reference camera poses. While structure-based methods have proved to deliver accurate camera pose estimates, they rely on pre-computed 3D descriptors coming from reference images often misaligned with query images. This discrepancy can subsequently harm downstream camera pose estimation tasks. In this paper we introduce the Feature Query Network (FQN), a ray-based descriptor regressor that can be used to query descriptors at known 3D locations under novel viewpoints. We show that the FQN is able to model viewpoint-dependency of high-dimensional keypoint descriptors and bring significant relative improvements to structure-based visual localization baselines.