VigilFormer: Deformable Attention for Video Anomaly Detection with Causal Risk Inference
Researchers have developed VigilFormer, a novel framework for video anomaly detection that balances accuracy with real-time processing. The system utilizes a Deformable Spatio-Temporal Encoder to efficiently focus on relevant video segments and a Causal Anomaly Classifier for distinguishing anomalies without frame-level labels. Additionally, an Adaptive Confidence Scheduler dynamically skips non-essential frames during inference to further optimize performance. AI