Lune

CVPR Findings2026

Disrupting Positional Encoding for Effective Open Set Recognition

Yu Wang, Jiabo Xie, Yucan Zhou, Junxian Mu, Qinghua Hu, Pengfei Zhu

Abstract

Open set recognition faces a fundamental challenge: accurately identifying known classes while effectively rejecting unknown samples. A key difficulty lies in generating suitable anomaly samples that lie near, yet outside, the distribution of known classes. This paper proposes a novel approach called ViPER, which disrupts the positional encodings in Vision Transformers (ViTs) to automatically generate highquality, semantically consistent near-Out-Of-Distribution (near-OOD) samples with structural anomalies. By preserving local image semantics while precisely interfering with the model's reliance on spatial structure, ViPER constructs "pseudo-outlier" samples near the decision boundaries of known classes in the feature space, providing powerful regularization signals. Theoretical analysis and experimental results demonstrate that ViPER significantly outperforms existing methods across multiple standard Open Set Recognition (OSR) tasks, achieving an average AUROC improvement of 3.4% (up to 5.6%) while maintaining excellent closed-set classification performance. Notably, ViPER is lightweight with no additional parameters or external data required. Code is available at ViPER.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Conditional Prompt Learning for Vision-Language Models cites Learning Transferable Visual Models From Natural Language Supervision
Builds onCited byLeft to right by year; the more cited, the higher.

Builds on24

  1. Learning Transferable Visual Models From Natural Language SupervisionAlec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh et al.ICML 2021 · 47,906 citations
  2. Conditional Prompt Learning for Vision-Language ModelsKaiyang Zhou, Jingkang Yang, Chen Change Loy, Ziwei LiuCVPR 2022 · 1,438 citations
  3. CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted InstancesJihoon Tack, Sangwoo Mo, Jongheon Jeong, Jinwoo ShinNeurIPS 2020 · 755 citations
  4. POEM: Out-of-Distribution Detection with Posterior SamplingYifei Ming, Ying Fan, Yixuan LiICML 2022 · 151 citations
  5. Open-set Label Noise Can Improve Robustness Against Inherent Label NoiseHongxin Wei, Lue Tao, Renchunzi Xie, Bo AnNeurIPS 2021 · 113 citations

Related papers

  1. DropPos: Pre-Training Vision Transformers by Reconstructing Dropped PositionsHaochen Wang, Junsong Fan, Yuxi Wang, Kaiyou Song et al.NeurIPS 2023 · 32 citations
  2. Exploring the Limits of Out-of-Distribution DetectionStanislav Fort, Jie Ren, Balaji LakshminarayananNeurIPS 2021 · 443 citations
  3. VisualAD: Language-Free Zero-Shot Anomaly Detection via Vision TransformerYanning Hou, Peiyuan Li, Zirui Liu, Yitong Wang et al.CVPR 2026 · 14 citations
  4. Delving Deep into the Generalization of Vision Transformers under Distribution ShiftsChongzhi Zhang, Mingyuan Zhang, Shanghang Zhang, Daisheng Jin et al.CVPR 2022 · 95 citations
  5. Sparsity as a Key: Unlocking New Insights from Latent Structures for Out-of-Distribution DetectionAhyoung Oh, Wonseok Shin, Songkuk KimCVPR 2026
Dusk over the sea between two cliffs drawn in fine vertical lines