Tracking the Discriminative Axis: Dual Prototypes for Test-Time OOD Detection Under Covariate Shift

• URL: http://arxiv.org/abs/2603.15213v2
• Date: Tue, 17 Mar 2026 07:50:17 GMT
• Title: Tracking the Discriminative Axis: Dual Prototypes for Test-Time OOD Detection Under Covariate Shift
• Abstract summary: Out-of-distribution (OOD) detection is indispensable for reliable deployment of deep-learning systems.<n>We propose DART, a test-time, online OOD detection method that dynamically tracks dual prototypes.<n>Our method significantly improves performance, yielding 15.32 percentage points (pp) AUROC gain and 49.15 pp FPR@95TPR reduction on ImageNet-C vs. Textures-C.
• Score: 9.525480587488095
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: For reliable deployment of deep-learning systems, out-of-distribution (OOD) detection is indispensable. In the real world, where test-time inputs often arrive as streaming mixtures of in-distribution (ID) and OOD samples under evolving covariate shifts, OOD samples are domain-constrained and bounded by the environment, and both ID and OOD are jointly affected by the same covariate factors. Existing methods typically assume a stationary ID distribution, but this assumption breaks down in such settings, leading to severe performance degradation. We empirically discover that, even under covariate shift, covariate-shifted ID (csID) and OOD (csOOD) samples remain separable along a discriminative axis in feature space. Building on this observation, we propose DART, a test-time, online OOD detection method that dynamically tracks dual prototypes -- one for ID and the other for OOD -- to recover the drifting discriminative axis, augmented with multi-layer fusion and flip correction for robustness. Extensive experiments on a wide range of challenging benchmarks, where all datasets are subjected to 15 common corruption types at severity level 5, demonstrate that our method significantly improves performance, yielding 15.32 percentage points (pp) AUROC gain and 49.15 pp FPR@95TPR reduction on ImageNet-C vs. Textures-C compared to established baselines. These results highlight the potential of the test-time discriminative axis tracking for dependable OOD detection in dynamically changing environments.

Related papers

• Mitigating the ID-OOD Tradeoff in Open-Set Test-Time Adaptation [52.620992698137236]
  Open-set test-time adaptation (OSTTA) addresses the challenge of adapting models to new environments.<n>In this paper, we first analyze the limitations of entropy in OSTTA and then introduce an angular loss to regulate feature norm magnitudes.<n>Our method achieves strong OOD detection while maintaining high ID classification performance.
  arXiv  Detail & Related papers  (2026-04-02T04:00:57Z)
• GOOD: Training-Free Guided Diffusion Sampling for Out-of-Distribution Detection [61.96025941146103]
  GOOD is a novel framework that guides sampling trajectories towards OOD regions using off-the-shelf in-distribution (ID) classifiers.<n> GOOD incorporates dual-level guidance: Image-level guidance based on the gradient of log partition to reduce input likelihood, drives samples toward low-density regions in pixel space.<n>We introduce a unified OOD score that adaptively combines image and feature discrepancies, enhancing detection robustness.
  arXiv  Detail & Related papers  (2025-10-20T03:58:46Z)
• Rethinking Out-of-Distribution Detection on Imbalanced Data Distribution [38.844580833635725]
  We present a training-time regularization technique to mitigate the bias and boost imbalanced OOD detectors across architecture designs. Our method translates into consistent improvements on the representative CIFAR10-LT, CIFAR100-LT, and ImageNet-LT benchmarks.
  arXiv  Detail & Related papers  (2024-07-23T12:28:59Z)
• Rethinking the Evaluation of Out-of-Distribution Detection: A Sorites Paradox [70.57120710151105]
  Most existing out-of-distribution (OOD) detection benchmarks classify samples with novel labels as the OOD data. Some marginal OOD samples actually have close semantic contents to the in-distribution (ID) sample, which makes determining the OOD sample a Sorites Paradox. We construct a benchmark named Incremental Shift OOD (IS-OOD) to address the issue.
  arXiv  Detail & Related papers  (2024-06-14T09:27:56Z)
• General-Purpose Multi-Modal OOD Detection Framework [5.287829685181842]
  Out-of-distribution (OOD) detection identifies test samples that differ from the training data, which is critical to ensuring the safety and reliability of machine learning (ML) systems. We propose a general-purpose weakly-supervised OOD detection framework, called WOOD, that combines a binary classifier and a contrastive learning component. We evaluate the proposed WOOD model on multiple real-world datasets, and the experimental results demonstrate that the WOOD model outperforms the state-of-the-art methods for multi-modal OOD detection.
  arXiv  Detail & Related papers  (2023-07-24T18:50:49Z)
• Supervision Adaptation Balancing In-distribution Generalization and Out-of-distribution Detection [36.66825830101456]
  In-distribution (ID) and out-of-distribution (OOD) samples can lead to textitdistributional vulnerability in deep neural networks. We introduce a novel textitsupervision adaptation approach to generate adaptive supervision information for OOD samples, making them more compatible with ID samples.
  arXiv  Detail & Related papers  (2022-06-19T11:16:44Z)
• How to Exploit Hyperspherical Embeddings for Out-of-Distribution Detection? [22.519572587827213]
  CIDER is a representation learning framework that exploits hyperspherical embeddings for OOD detection. CIDER establishes superior performance, outperforming the latest rival by 19.36% in FPR95.
  arXiv  Detail & Related papers  (2022-03-08T23:44:01Z)
• Exploring Covariate and Concept Shift for Detection and Calibration of Out-of-Distribution Data [77.27338842609153]
  characterization reveals that sensitivity to each type of shift is important to the detection and confidence calibration of OOD data. We propose a geometrically-inspired method to improve OOD detection under both shifts with only in-distribution data. We are the first to propose a method that works well across both OOD detection and calibration and under different types of shifts.
  arXiv  Detail & Related papers  (2021-10-28T15:42:55Z)
• MOOD: Multi-level Out-of-distribution Detection [13.207044902083057]
  Out-of-distribution (OOD) detection is essential to prevent anomalous inputs from causing a model to fail during deployment. We propose a novel framework, multi-level out-of-distribution detection MOOD, which exploits intermediate classifier outputs for dynamic and efficient OOD inference. MOOD achieves up to 71.05% computational reduction in inference, while maintaining competitive OOD detection performance.
  arXiv  Detail & Related papers  (2021-04-30T02:18:31Z)
• Learn what you can't learn: Regularized Ensembles for Transductive Out-of-distribution Detection [76.39067237772286]
  We show that current out-of-distribution (OOD) detection algorithms for neural networks produce unsatisfactory results in a variety of OOD detection scenarios. This paper studies how such "hard" OOD scenarios can benefit from adjusting the detection method after observing a batch of the test data. We propose a novel method that uses an artificial labeling scheme for the test data and regularization to obtain ensembles of models that produce contradictory predictions only on the OOD samples in a test batch.
  arXiv  Detail & Related papers  (2020-12-10T16:55:13Z)
• Robust Out-of-distribution Detection for Neural Networks [51.19164318924997]
  We show that existing detection mechanisms can be extremely brittle when evaluating on in-distribution and OOD inputs. We propose an effective algorithm called ALOE, which performs robust training by exposing the model to both adversarially crafted inlier and outlier examples.
  arXiv  Detail & Related papers  (2020-03-21T17:46:28Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.