サインイン
インサイト - Visual place recognition
A Lightweight Teacher-Student Model with Cross-Metric Knowledge Distillation for Efficient Visual Place Recognition
A novel cross-metric knowledge distillation approach that enables a lightweight student model to outperform a more complex teacher model in visual place recognition tasks, while maintaining superior performance and computational efficiency.
Estimating Image-Matching Uncertainty for Robust Visual Place Recognition
Reliable uncertainty estimation is key to avoid catastrophic failures in visual place recognition pipelines due to perceptual aliasing. This work compares three main categories of uncertainty estimation methods and proposes a simple baseline that considers the spatial locations of the reference images.