Quantifying Epistemic Uncertainty in Absolute Pose Regression

2025-04-09

AI Summary

Absolute pose regression models suffer from unreliable predictions and a lack of calibrated epistemic uncertainty quantification when deployed in out-of-distribution scenes. To address this, we propose the first variational inference-based observation likelihood modeling framework for pose regression, formulating camera localization as a probabilistic inference problem. Our method jointly resolves observation ambiguities arising from repetitive structures and enables interpretable, probabilistic pose estimation. Within a Bayesian deep learning framework, we explicitly model epistemic uncertainty in pose predictions, substantially improving the correlation between estimated uncertainty and actual pose error. Evaluated on multiple visual relocalization benchmarks—including 7Scenes and Cambridge Landmarks—our approach achieves state-of-the-art performance in both uncertainty calibration and pose accuracy, as measured by their joint metric.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Intelligent Robots: State Estimation

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies

Abstract

Visual relocalization is the task of estimating the camera pose given an image it views. Absolute pose regression offers a solution to this task by training a neural network, directly regressing the camera pose from image features. While an attractive solution in terms of memory and compute efficiency, absolute pose regression's predictions are inaccurate and unreliable outside the training domain. In this work, we propose a novel method for quantifying the epistemic uncertainty of an absolute pose regression model by estimating the likelihood of observations within a variational framework. Beyond providing a measure of confidence in predictions, our approach offers a unified model that also handles observation ambiguities, probabilistically localizing the camera in the presence of repetitive structures. Our method outperforms existing approaches in capturing the relation between uncertainty and prediction error.

Problem

• Quantify epistemic uncertainty in absolute pose regression
• Improve camera pose prediction accuracy outside training domain
• Handle observation ambiguities in repetitive structures

Innovation

• Estimates epistemic uncertainty via variational framework
• Unified model handles observation ambiguities probabilistically
• Improves uncertainty-prediction error correlation accuracy