The Flow has not summarised this story yet — read it at arXiv Computer Vision.
The paper investigates how weakly supervised video anomaly detectors, trained with only video‑level labels, are evaluated using frame‑level metrics such as Micro‑AUROC and AP. It shows that these metrics largely measure a detector’s ability to separate different videos rather than correctly ordering anomalous moments within a single video, a phenomenon termed temporal dilution. Experiments demonstrate that a detector can achieve high pooled scores even when it assigns the same score to every frame in a video, indicating that current evaluation practices may overstate temporal localization performance.
The paper introduces an adaptive temporal modeling framework for weakly supervised video anomaly detection that addresses the limitations of rigid Multiple Instance Learning approaches. It presents a Temporal Refinement Module using dynamic positional encoding and a learnable class token to capture long‑range dependencies, and an Event Segmentation Module that identifies event boundaries via temporal discontinuity analysis to produce discriminative event‑level representations. An adaptive similarity‑based fusion strategy replaces fixed top‑k heuristics, dynamically integrating snippet‑level and event‑level anomaly scores into video‑level predictions, and the method outperforms state‑of‑the‑art baselines on two benchmarks.
The paper investigates whether detailed articulated human pose provides more discriminative power than coarse spatial relationships for early violence detection. By fixing the downstream pipeline and comparing five interaction representations—including bounding‑box geometry, handcrafted pose analogues, enriched pose descriptors, and a learned joint encoder—the study finds that pose‑based representations do not outperform coarse geometry. When visual encoders are frozen and evaluated on larger datasets, person‑crop appearance and whole‑frame context outperform geometry, but cropping to interacting people offers no advantage over encoding the entire frame. The authors further demonstrate that pre‑onset frames contain source‑related artifacts (e.g., title cards, watermarks) that contribute significantly to discrimination, suggesting that benchmark performance may reflect these artifacts rather than true event evidence.
Video anomaly detection (VAD) aims to identify and temporally localize abnormal events in videos. Supervised methods learn anomaly decision boundaries from target-domain annotations but require substantial in-domain data.
Weakly supervised video anomaly detection (WSVAD) has predominantly focused on temporal localization, identifying when anomalies occur while largely neglecting their spatial extent within frames. Yet, spatial localization is essential for interpretability and practical deployment in real-world settings.
The paper proposes a lightweight federated multiple‑instance learning (MIL) framework that trains only a compact MIL scorer across distributed clients while using a frozen vision‑language model (VLM) to verify high‑scoring video segments post‑hoc. Two VLM feedback interfaces are explored: a parsed text‑generation interface and a logit‑based interface that derives a continuous anomaly score from next‑token Yes/No probabilities. Experiments on UCF‑Crime with InternVL3.5‑2B and Qwen3‑VL‑2B‑Instruct show that the logit interface consistently improves frame‑level AUC and AP over the MIL baseline without requiring temporal post‑processing, whereas the text‑generation interface is more sensitive to prompts, parsers, and model choice.