Anomaly Amplification in Videos Using F2LM Generator and Destroyer
D C Subhashree, Attulaya Kumar Singh, Ayush, Adesh R Jain, Aditya Pansari · 2025
Seungkyun Hong, Sunghyun Ahn, Youngwan Jo, and Sanghyun Park introduced the Generator-Destroyer framework for video anomaly detection in 2024. This architecture employs a generator to predict future frames and a destroyer to highlight anomalies by omitting degraded regions. While effective in isolating anomalies, this study enhances the original system by integrating a Gaussian filtering stage during preprocessing. This additional step reduces high-frequency noise, improving the system’s ability to localize anomalies. As a result, the enhanced framework shows improved detection accuracy, reduced false positives, and faster computation on benchmark datasets. These improvements significantly boost the system’s robustness and reliability, particularly in noisy visual environments.