Anomaly classification task

Anomaly classification task

Anomaly classification is a semi-supervised task of finding patterns that considerably deviate from standard patterns. When you apply the method of anomaly classification, you would first feed the model with a dataset of normal patterns. After the initial training on the dataset with regular patterns, the second round of training would include a dataset with anomalous patterns. After this operation, the model can differentiate “anomalous” from “normal” since the normal patterns appear more often than abnormal patterns. In this task, you may be required to classify images with only regular patterns.

The Intel® Geti™ platform supports 3 anomaly task types:

• Anomaly classification

• Anomaly detection

• Anomaly segmentation

When you create an anomaly classification project, a classifier labels the whole image either as “normal” or “anomalous”. In anomaly detection, the classifier localizes the anomaly and draws a bounding box around the anomalous object. In anomaly segmentation, the classifier localizes the anomaly and draws a polygon around the anomalous object.