Self-Distilled Masked Auto-Encoders are Efficient Video Anomaly Detectors

TL;DR

An efficient abnormal event detection model based on a lightweight masked auto-encoder applied at the video frame level that achieves an excellent trade-off between speed and accuracy, obtaining competitive AUC scores, while processing 1655 FPS is proposed.

Abstract

We propose an efficient abnormal event detection model based on a lightweight masked auto-encoder (AE) applied at the video frame level. The novelty of the proposed model is threefold. First, we introduce an approach to weight tokens based on motion gradients, thus shifting the focus from the static background scene to the foreground objects. Second, we integrate a teacher decoder and a student decoder into our architecture, leveraging the discrepancy between the outputs given by the two decoders to improve anomaly detection. Third, we generate synthetic abnormal events to augment the training videos, and task the masked AE model to jointly reconstruct the original frames (without anomalies) and the corresponding pixel-level anomaly maps. Our design leads to an efficient and effective model, as demonstrated by the extensive experiments carried out on four benchmarks: Avenue, ShanghaiTech, UBnormal and UCSD Ped2. The empirical results show that our model achieves an excellent trade-off between speed and accuracy, obtaining competitive AUC scores, while processing 1655 FPS. Hence, our model is between 8 and 70 times faster than competing methods. We also conduct an ablation study to justify our design. Our code is freely available at: https://github.com/ristea/aed-mae.

Figures (12)

• Figure 1: Our masked auto-encoder for abnormal event detection based on self-distillation. At training time, some video frames are augmented with synthetic anomalies. The teacher decoder learns to reconstruct original frames (without anomalies) and predict anomaly maps. The student decoder learns to reproduce the teacher's output. Motion gradients are aggregated at the token level and used as weights for the reconstruction loss. Red dashed lines represent steps executed only during training.
• Figure 2: Performance versus speed trade-offs for our self-distilled masked AE and several state-of-the-art methods Georgescu-CVPR-2021Georgescu-TPAMI-2021Ristea-CVPR-2022Liu-CVPR-2018Liu-ICCV-2021Park-CVPR-2020Gong-ICCV-2019Wang-ECCV-2022Park-WACV-2022 (with open-sourced code), on the Avenue data set. The running times of all methods are measured on a computer with one Nvidia GeForce GTX 3090 GPU with 24 GB of VRAM. Best viewed in color.
• Figure 3: Four synthetic anomalies (with red contours) taken from the UBnormal data set Acsintoae-CVPR-2022 and overlaid on training frames from Avenue Lu-ICCV-2013. Best viewed in color.
• Figure 4: Predictions for test video $04$ from Avenue. The abnormal bounding boxes are given by the convex hull of the patches labeled as abnormal. Best viewed in color.
• Figure 5: Examples of frames and anomaly maps reconstructed by our teacher. The first four columns correspond to abnormal examples from the Avenue and ShanghaiTech data sets, while the last column shows a normal example. Best viewed in color.