STEAD: Spatio-Temporal Efficient Anomaly Detection for Time and Compute Sensitive Applications

This paper presents a new method for anomaly detection in automated systemswith time and compute sensitive requirements, such as autonomous driving, withunparalleled efficiency. As systems like autonomous driving become increasinglypopular, ensuring their safety has become more important than ever. Therefore,this paper focuses on how to quickly and effectively detect various anomaliesin the aforementioned systems, with the goal of making them safer and moreeffective. Many detection systems have been developed with great success underspatial contexts; however, there is still significant room for improvement whenit comes to temporal context. While there is substantial work regarding thistask, there is minimal work done regarding the efficiency of models and theirability to be applied to scenarios that require real-time inference, i.e.,autonomous driving where anomalies need to be detected the moment they arewithin view. To address this gap, we propose STEAD (Spatio-Temporal EfficientAnomaly Detection), whose backbone is developed using (2+1)D Convolutions andPerformer Linear Attention, which ensures computational efficiency withoutsacrificing performance. When tested on the UCF-Crime benchmark, our base modelachieves an AUC of 91.34%, outperforming the previous state-of-the-art, and ourfast version achieves an AUC of 88.87%, while having 99.70% less parameters andoutperforming the previous state-of-the-art as well. The code and pretrainedmodels are made publicly available at https://github.com/agao8/STEAD