A Geometry-Based View of Mahalanobis OOD Detection

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Janiak, Denis, Binkowski, Jakub, Kajdanowicz, Tomasz
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911483310374912
author Janiak, Denis
Binkowski, Jakub
Kajdanowicz, Tomasz
author_facet Janiak, Denis
Binkowski, Jakub
Kajdanowicz, Tomasz
contents Out-of-distribution (OOD) detection is critical for reliable deployment of vision models. Mahalanobis-based detectors remain strong baselines, yet their performance varies widely across modern pretrained representations, and it is unclear which properties of a feature space cause these methods to succeed or fail. We conduct a large-scale study across diverse foundation-model backbones and Mahalanobis variants. First, we show that Mahalanobis-style OOD detection is not universally reliable: performance is highly representation-dependent and can shift substantially with pretraining data and fine-tuning regimes. Second, we link this variability to in-distribution geometry and identify a two-term ID summary that consistently tracks Mahalanobis OOD behavior across detectors: within-class spectral structure and local intrinsic dimensionality. Finally, we treat normalization as a geometric control mechanism and introduce radially scaled $\ell_2$ normalization, $ϕ_β(z)=z/\|z\|^β$, which preserves directions while contracting or expanding feature radii. Varying $β$ changes the radii while preserving directions, so the same quadratic detector sees a different ID geometry. We choose $β$ from ID-only geometry signals and typically outperform fixed normalization baselines.
format Preprint
id arxiv_https___arxiv_org_abs_2510_15202
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Geometry-Based View of Mahalanobis OOD Detection
Janiak, Denis
Binkowski, Jakub
Kajdanowicz, Tomasz
Machine Learning
Computer Vision and Pattern Recognition
Out-of-distribution (OOD) detection is critical for reliable deployment of vision models. Mahalanobis-based detectors remain strong baselines, yet their performance varies widely across modern pretrained representations, and it is unclear which properties of a feature space cause these methods to succeed or fail. We conduct a large-scale study across diverse foundation-model backbones and Mahalanobis variants. First, we show that Mahalanobis-style OOD detection is not universally reliable: performance is highly representation-dependent and can shift substantially with pretraining data and fine-tuning regimes. Second, we link this variability to in-distribution geometry and identify a two-term ID summary that consistently tracks Mahalanobis OOD behavior across detectors: within-class spectral structure and local intrinsic dimensionality. Finally, we treat normalization as a geometric control mechanism and introduce radially scaled $\ell_2$ normalization, $ϕ_β(z)=z/\|z\|^β$, which preserves directions while contracting or expanding feature radii. Varying $β$ changes the radii while preserving directions, so the same quadratic detector sees a different ID geometry. We choose $β$ from ID-only geometry signals and typically outperform fixed normalization baselines.
title A Geometry-Based View of Mahalanobis OOD Detection
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2510.15202