Learning to Localize in Unseen Scenes with Relative Pose Regressors

TL;DR

This paper introduces a novel approach for relative pose regression that improves generalization to unseen scenes by aggregating feature maps into latent codes and using advanced rotation representations, outperforming existing methods.

Contribution

The authors propose a new RPR architecture that aggregates features into latent codes and employs Transformer Encoders and continuous rotation representations for better unseen scene localization.

Findings

Significantly improved localization in unseen environments.

Maintains competitive performance in seen scenes.

Validated through multiple ablation studies.

Abstract

Relative pose regressors (RPRs) localize a camera by estimating its relative translation and rotation to a pose-labelled reference. Unlike scene coordinate regression and absolute pose regression methods, which learn absolute scene parameters, RPRs can (theoretically) localize in unseen environments, since they only learn the residual pose between camera pairs. In practice, however, the performance of RPRs is significantly degraded in unseen scenes. In this work, we propose to aggregate paired feature maps into latent codes, instead of operating on global image descriptors, in order to improve the generalization of RPRs. We implement aggregation with concatenation, projection, and attention operations (Transformer Encoders) and learn to regress the relative pose parameters from the resulting latent codes. We further make use of a recently proposed continuous representation of rotation matrices, which alleviates the limitations of the commonly used quaternions. Compared to state-of-the-art RPRs, our model is shown to localize significantly better in unseen environments, across both indoor and outdoor benchmarks, while maintaining competitive performance in seen scenes. We validate our findings and architecture design through multiple ablations. Our code and pretrained models is publicly available.