mlsanityData to 7 Oct 2026

NOVA: Normal-Side Modeling for Training-Free Zero-Shot Video Anomaly Detection

arXiv 2609.06360 ↗First posted 6 Sep 2026cs.CVcited by 0

On the map: Multimodal LLMs / vision-language models → Chart understanding and anomaly detection

Abstract (arXiv)

Training-free zero-shot video anomaly detection (ZS-VAD) leverages vision-language models (VLMs) to localize anomaly instances from a predefined anomaly vocabulary, without providing any video. Existing CLIP-based methods often emphasize anomaly-side semantics, while the competing normality side remains less carefully formulated. We identify two key limitations in existing solutions: (i) blurred decision boundary: normal prompts may contain ambiguous verbs, such as running, that are semantically close to anomalies, reducing normal and abnormal separation in the VLM embedding space; and (ii) modality gap: poor alignment between features of textual normal anchors and visual frames. We propose NOVA, a training-free ZS-VAD framework that strengthens the normal side at both linguistic and visual levels. NOVA introduces Normality-Aware Prompt Construction (NA), which excludes anomaly-adjacent verbs and biases normal descriptions toward static, low-motion scenes. To overcome the text-vision modality gap, NOVA constructs a Visual Normality Anchor (VNA), which creates a weighted visual normal anchor from the initial frames of each test video, providing a video-specific normal reference without task-specific training or annotations. NOVA achieves 89.86 percent AUC on UCF-Crime and 95.07 percent AUC and 84.82 percent AP on XD-Violence, reaching state-of-the-art performance among comparable training-free zero-shot methods.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Chart understanding and anomaly detection, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 30 of its 41 references that are in the map. First posted on arXiv 6 Sep 2026.

Builds on

30 of 41 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025Cerberus: Real-Time Video Anomaly Detection via Cascaded Vision-Language Models18 Oct
  2. 2025Unlocking Vision-Language Models for Video Anomaly Detection via Fine-Grained Prompting2 Oct
  3. 2024Holmes-VAD: Towards Unbiased and Explainable Video Anomaly Detection via Multi-modal LLM18 Jun
  4. 2023VadCLIP: Adapting Vision-Language Models for Weakly Supervised Video Anomaly Detection22 Aug

Other references · 26

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  2. 2021Learning to Prompt for Vision-Language Modelscited by 2,269
  3. 2018Real-world Anomaly Detection in Surveillance Videoscited by 570
  4. 2022Mind the Gap: Understanding the Modality Gap in Multi-modal Contrastive Representation Learningcited by 553
  5. 2025Perception Encoder: The best visual embeddings are not at the output of the networkcited by 315
  6. 2023WinCLIP: Zero-/Few-Shot Anomaly Classification and Segmentationcited by 311
  7. 2023AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detectioncited by 270
  8. 2022What does a platypus look like? Generating customized prompts for zero-shot image classificationcited by 263
  9. 2020Not only Look, but also Listen: Learning Multimodal Violence Detection under Weak Supervisioncited by 235
  10. 2021Weakly-supervised Video Anomaly Detection with Robust Temporal Feature Magnitude Learningcited by 226
  11. 2022Generative Cooperative Learning for Unsupervised Video Anomaly Detectioncited by 102
  12. 2020A Survey of Single-Scene Video Anomaly Detectioncited by 88
14 more
  1. 2024Harnessing Large Language Models for Training-free Video Anomaly Detectioncited by 85
  2. 2022CLIP-TSA: CLIP-Assisted Temporal Self-Attention for Weakly-Supervised Video Anomaly Detectioncited by 67
  3. 2019GODS: Generalized One-class Discriminative Subspaces for Anomaly Detectioncited by 64
  4. 2025Exploring Intrinsic Normal Prototypes within a Single Image for Universal Anomaly Detectioncited by 47
  5. 2025EventVAD: Training-Free Event-Aware Video Anomaly Detectioncited by 32
  6. 2022DyAnNet: A Scene Dynamicity Guided Self-Trained Video Anomaly Detection Networkcited by 26
  7. 2024MULDE: Multiscale Log-Density Estimation via Denoising Score Matching for Video Anomaly Detectioncited by 21
  8. 2025AnyAnomaly: Zero-Shot Customizable Video Anomaly Detection with LVLMcited by 15
  9. 2025PANDA: Towards Generalist Video Anomaly Detection via Agentic AI Engineercited by 15
  10. 2025VADTree: Explainable Training-Free Video Anomaly Detection via Hierarchical Granularity-Aware Treecited by 15
  11. 2025Anomize: Better Open Vocabulary Video Anomaly Detectioncited by 13
  12. 2025Flashback: Memory-Driven Zero-shot, Real-time Video Anomaly Detectioncited by 12
  13. 2025Language-guided Open-world Video Anomaly Detection under Weak Supervisioncited by 8
  14. 2026No Need For Real Anomaly: MLLM Empowered Zero-Shot Video Anomaly Detectioncited by 6

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2024Do LLMs Understand Visual Anomalies? Uncovering LLM's Capabilities in Zero-shot Anomaly Detection
  2. 2026VisualAD: Language-Free Zero-Shot Anomaly Detection via Vision Transformer
  3. 2026Is Video Anomaly Detection Misframed? Evidence from LLM-Based and Multi-Scene Models
  4. 2026Probe-VAD: Ordinal Likelihood Probing for Training-Free Video Anomaly Detection
  5. 2026PrismVAU: Prompt-Refined Inference System for Multimodal Video Anomaly Understanding
  6. 2024VCP-CLIP: A visual context prompting model for zero-shot anomaly segmentation
  7. 2026Are Multimodal LLMs Ready for Surveillance? A Reality Check on Zero-Shot Anomaly Detection in the Wild
  8. 2026CoReVAD: A Contextual Reasoning Framework for Training-Free Video Anomaly Detection