Uncertainty-Aware Knowledge Distillation for Compact and Efficient 6DoF Pose Estimation

2025-03-17

AI Summary

To address the demand for lightweight 6DoF pose estimation models in real-time applications such as robotics, AR, and autonomous space navigation, this paper proposes an uncertainty-aware knowledge distillation framework. Methodologically, it introduces a novel dynamically weighted distillation strategy guided by teacher-predicted keypoint uncertainties, coupled with uncertainty-driven feature map localization for targeted knowledge transfer. Integrating uncertainty modeling, end-to-end distillation, and lightweight network design, the approach achieves state-of-the-art performance on LINEMOD and SPEED+ datasets: it attains higher pose accuracy (ADD(-S) improved by 2.1–4.8 percentage points) with a significantly smaller model (37% fewer parameters) and demonstrates strong cross-domain generalization. The core contribution lies in explicitly modeling prediction uncertainty as both distillation weights and spatial masks for feature-level knowledge transfer—enabling, for the first time, uncertainty-guided distillation of 6DoF keypoint representations.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Intelligent Robots: State Estimation
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Compact and efficient 6DoF object pose estimation is crucial in applications such as robotics, augmented reality, and space autonomous navigation systems, where lightweight models are critical for real-time accurate performance. This paper introduces a novel uncertainty-aware end-to-end Knowledge Distillation (KD) framework focused on keypoint-based 6DoF pose estimation. Keypoints predicted by a large teacher model exhibit varying levels of uncertainty that can be exploited within the distillation process to enhance the accuracy of the student model while ensuring its compactness. To this end, we propose a distillation strategy that aligns the student and teacher predictions by adjusting the knowledge transfer based on the uncertainty associated with each teacher keypoint prediction. Additionally, the proposed KD leverages this uncertainty-aware alignment of keypoints to transfer the knowledge at key locations of their respective feature maps. Experiments on the widely-used LINEMOD benchmark demonstrate the effectiveness of our method, achieving superior 6DoF object pose estimation with lightweight models compared to state-of-the-art approaches. Further validation on the SPEED+ dataset for spacecraft pose estimation highlights the robustness of our approach under diverse 6DoF pose estimation scenarios.

Problem

• Develops uncertainty-aware knowledge distillation for 6DoF pose estimation.
• Enhances student model accuracy using teacher model uncertainty.
• Achieves lightweight, real-time 6DoF pose estimation in robotics and AR.

Innovation

• Uncertainty-aware Knowledge Distillation for 6DoF pose estimation
• Keypoint-based alignment using teacher model uncertainty
• Lightweight models achieve superior pose estimation accuracy