NM-GAN: Noise-modulated generative adversarial network for video anomaly detection

作者：

Highlights：

• A more accurate and stable model for video anomaly detection is achieved within a refined end-to-end GAN-like architecture.

• The reconstruction network has stronger and more controllable generalization ability.

• The discrimination network uses the reconstruction error map to distinguish anomaly samples.

• The proposed noise-modulated adversarial learning method enhances the ability of the discriminator to detect anomalies.

摘要

•A more accurate and stable model for video anomaly detection is achieved within a refined end-to-end GAN-like architecture.•The reconstruction network has stronger and more controllable generalization ability.•The discrimination network uses the reconstruction error map to distinguish anomaly samples.•The proposed noise-modulated adversarial learning method enhances the ability of the discriminator to detect anomalies.

论文关键词：Video anomaly detection,Generative adversarial network,Noise modulation,Reconstruction error map,Generalization ability

论文评审过程：Received 22 April 2020, Revised 20 November 2020, Accepted 26 March 2021, Available online 1 April 2021, Version of Record 11 April 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.107969