Benchmarking Compact VLMs for Clip-Level Surveillance Anomaly Detection Under Weak Supervision

TL;DR

This paper evaluates compact vision-language models for clip-level surveillance anomaly detection, demonstrating that parameter-efficient fine-tuning achieves high accuracy and efficiency, outperforming some baselines under a standardized evaluation protocol.

Contribution

It introduces a unified evaluation protocol and shows that compact VLMs with parameter-efficient adaptation can serve as reliable, efficient anomaly detectors in surveillance settings.

Findings

Compact VLMs achieve comparable or better performance than baselines.

Parameter-efficient adaptation reduces prompt sensitivity.

Compact VLMs maintain low latency while ensuring detection accuracy.

Abstract

CCTV safety monitoring demands anomaly detectors combine reliable clip-level accuracy with predictable per-clip latency despite weak supervision. This work investigates compact vision-language models (VLMs) as practical detectors for this regime. A unified evaluation protocol standardizes preprocessing, prompting, dataset splits, metrics, and runtime settings to compare parameter-efficiently adapted compact VLMs against training-free VLM pipelines and weakly supervised baselines. Evaluation spans accuracy, precision, recall, F1, ROC-AUC, and average per-clip latency to jointly quantify detection quality and efficiency. With parameter-efficient adaptation, compact VLMs achieve performance on par with, and in several cases exceeding, established approaches while retaining competitive per-clip latency. Adaptation further reduces prompt sensitivity, producing more consistent behavior across prompt regimes under the shared protocol. These results show that parameter-efficient fine-tuning enables compact VLMs to serve as dependable clip-level anomaly detectors, yielding a favorable accuracy-efficiency trade-off within a transparent and consistent experimental setup.