NeFut Logo NeFut
Admin Login

[CS.AI] VigilFormer: Breakthrough in Video Anomaly Detection with Deformable Attention

Published at: 2026-06-17 22:00 Last updated: 2026-06-20 13:45
#algorithm #AI #Machine Learning

In surveillance settings, video anomaly detection must balance detection accuracy against real-time throughput. Existing methods typically address this tension through stronger feature extractors or more efficient architectures, but rarely both. We present VigilFormer, a unified framework that combines deformable spatio-temporal attention with causal temporal modeling to detect anomalies in untrimmed surveillance video.

The proposed Deformable Spatio-Temporal Encoder (DSTE) attends to a sparse set of informative locations across frames, avoiding the quadratic cost of dense attention while retaining the ability to capture irregular motion patterns. A Causal Anomaly Classifier (CAC) applies dilated causal convolutions over snippet-level features and optimizes a contrastive multiple-instance learning objective that separates anomalous and normal representations without frame-level labels. To meet deployment constraints, an Adaptive Confidence Scheduler (ACS) dynamically skips low-information frames at inference time, reducing redundant computation in static scenes.

Evaluated on UCF-Crime, ShanghaiTech, and CUHK Avenue, VigilFormer achieves AUC scores of 87.83%, 97.21%, and 89.74% respectively, at 41.5 FPS on a single GPU, outperforming recent weakly-supervised methods in both accuracy and speed.

Blogger's Review: VigilFormer successfully addresses the efficiency-accuracy trade-off in video anomaly detection by integrating deformable spatio-temporal attention and causal modeling. This innovative approach not only presents a new solution for surveillance video analysis but also demonstrates significant practical value. Its impressive performance across various datasets highlights its potential in real-world applications and sets a new direction for future research.

Original Source: https://arxiv.org/abs/2606.14724

[h] Back to Home