Hyperbolic Metric Learning for Visual Outlier Detection

arXiv:2403.15260v1 Announce Type: new
Abstract: Out-Of-Distribution (OOD) detection is critical to deploy deep learning models in safety-critical applications. However, the inherent hierarchical concept structure of visual data, which is instrumental to OOD detection, is often poorly captured by conventional methods based on Euclidean geometry. This work proposes a metric framework that leverages the strengths of Hyperbolic geometry for OOD detection. Inspired by previous works that refine the decision boundary for OOD data with synthetic outliers, we extend this method to …

abstract applications arxiv concept cs.cv data deep learning deploy detection distribution framework geometry hierarchical however outlier safety safety-critical type visual visual data work