# Auditing Frame-Level AUC in Weakly Supervised Video Anomaly Detection: Granularity, Resolution, and Scene Bias

arXiv: https://arxiv.org/abs/2608.11985 · HTML: https://mlsanity.com/paper/2608.11985/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 20 of its 32 references that are in the map.
- First posted on arXiv 12 Aug 2026.

## Builds on (most cited first, 20 of 20)

- [Beyond Accuracy: Behavioral Testing of NLP models with CheckList](https://mlsanity.com/paper/2005.04118/) (2020)
- [Real-world Anomaly Detection in Surveillance Videos](https://mlsanity.com/paper/1801.04264/) (2018)
- [Domino: Discovering Systematic Errors with Cross-Modal Embeddings](https://mlsanity.com/paper/2203.14960/) (2022)
- [VadCLIP: Adapting Vision-Language Models for Weakly Supervised Video Anomaly Detection](https://mlsanity.com/paper/2308.11681/) (2023)
- [Street Scene: A new dataset and evaluation protocol for video anomaly detection](https://mlsanity.com/paper/1902.05872/) (2019)
- [Dual Memory Units with Uncertainty Regulation for Weakly Supervised Video Anomaly Detection](https://mlsanity.com/paper/2302.05160/) (2023)
- [The Spotlight: A General Method for Discovering Systematic Errors in Deep Learning Models](https://mlsanity.com/paper/2107.00758/) (2021)
- [A Closer Look at AUROC and AUPRC under Class Imbalance](https://mlsanity.com/paper/2401.06091/) (2024)
- [Advancing Video Anomaly Detection: A Concise Review and a New Dataset](https://mlsanity.com/paper/2402.04857/) (2024)
- [Learning Prompt-Enhanced Context Features for Weakly-Supervised Video Anomaly Detection](https://mlsanity.com/paper/2306.14451/) (2023)
- [Weakly Supervised Video Anomaly Detection and Localization with Spatio-Temporal Prompts](https://mlsanity.com/paper/2408.05905/) (2024)
- [BatchNorm-based Weakly Supervised Video Anomaly Detection](https://mlsanity.com/paper/2311.15367/) (2023)
- [VGGSounder: Audio-Visual Evaluations for Foundation Models](https://mlsanity.com/paper/2508.08237/) (2025)
- [Learning to Tell Apart: Weakly Supervised Video Anomaly Detection via Disentangled Semantic Alignment](https://mlsanity.com/paper/2511.10334/) (2025)
- [Area under the ROC Curve has the Most Consistent Evaluation for Binary Classification](https://mlsanity.com/paper/2408.10193/) (2024)
- [Mixture of Experts Guided by Gaussian Splatters Matters: A new Approach to Weakly-Supervised Video Anomaly Detection](https://mlsanity.com/paper/2508.06318/) (2025)
- [Rethinking Metrics and Benchmarks of Video Anomaly Detection](https://mlsanity.com/paper/2505.19022/) (2025)
- [Is AUC the best measure for practical comparison of anomaly detectors?](https://mlsanity.com/paper/2305.04754/) (2023)
- [From Frames to Events: Rethinking Evaluation in Human-Centric Video Anomaly Detection](https://mlsanity.com/paper/2604.09327/) (2026)
- [Is Video Anomaly Detection Misframed? Evidence from LLM-Based and Multi-Scene Models](https://mlsanity.com/paper/2605.12725/) (2026)

## Built on by (most cited first, 1 of 1)

- [What Does an Evaluation License? A Commit-Bound Census of Claim Replay in Inspect Evals](https://mlsanity.com/paper/2608.19269/) (2026)

## Abstract (arXiv)

Frame-level area under the ROC curve (AUC) is the dominant evaluation metric for weakly supervised video anomaly detection (WSVAD). Its standard form measures whether an anomalous frame outranks a normal frame drawn from anywhere in the test set. We refer to this comparison as pooled AUC, since it aggregates frame pairs across test videos regardless of source. Pooled AUC therefore credits both event localization and differences between video sources. We audit this protocol on UCF-Crime across recent state-of-the-art models spanning different backbone families. Holding each model's frame scores fixed, we read them under three pairing granularities: global, per anomaly category, and within each video, then repeat the same three-granularity readout on zero-shot scores computed from the models' internal representations. We assess ranking reliability with a paired video bootstrap. Three findings follow. First, pooled AUC does not reliably predict within-video anomaly localization: models with similar pooled scores exhibit large localization differences and rank reversals under stricter granularities. Second, at the benchmark's test-split size, pooled AUC lacks the resolution to support state-of-the-art margins reported in the field. Within each backbone family, it resolves no comparison at those margins, while within-video AUC resolves several over identical predictions. Learned representations further reveal that within-video anomaly structure and detector localization are decoupled. Third, on normal footage alone, every model we examine separates videos by recording properties, such as resolution and color encoding, indicating that scene sensitivity is shared across the setting rather than specific to any architecture. We publicly release a granularity-aware protocol computable from existing predictions and scene-factor annotations for UCF-Crime.