eduzhai > Applied Sciences > Engineering >

Self-Supervised Representation Learning for Visual Anomaly Detection

  • Save

... pages left unread,continue reading

Document pages: 9 pages

Abstract: Self-supervised learning allows for better utilization of unlabelled data.The feature representation obtained by self-supervision can be used indownstream tasks such as classification, object detection, segmentation, andanomaly detection. While classification, object detection, and segmentationhave been investigated with self-supervised learning, anomaly detection needsmore attention. We consider the problem of anomaly detection in images andvideos, and present a new visual anomaly detection technique for videos.Numerous seminal and state-of-the-art self-supervised methods are evaluated foranomaly detection on a variety of image datasets. The best performingimage-based self-supervised representation learning method is then used forvideo anomaly detection to see the importance of spatial features in visualanomaly detection in videos. We also propose a simple self-supervision approachfor learning temporal coherence across video frames without the use of anyoptical flow information. At its core, our method identifies the frame indicesof a jumbled video sequence allowing it to learn the spatiotemporal features ofthe video. This intuitive approach shows superior performance of visual anomalydetection compared to numerous methods for images and videos on UCF101 andILSVRC2015 video datasets.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...