Computer Vision Meetup: Anomaly Detection with Anomalib and FiftyOne

Most anomaly detection techniques are unsupervised, meaning that anomaly detection models are trained on unlabeled non-anomalous data. Developing the highest-quality dataset and data pipeline is essential to training robust anomaly detection models.

In this brief walkthrough, I will illustrate how to leverage open-source FiftyOne and Anomalib to build deployment-ready anomaly detection models. First, we will load and visualize the MVTec AD dataset in the FiftyOne App. Next, we will use Albumentations to test out augmentation techniques. We will then train …

ai anomaly anomaly detection build computer computer vision computervision data data pipeline dataset deployment detection machinelearning meaning meetup pipeline quality robotics robust training unsupervised vision walkthrough will