Finding Outliers

Use Encord Active to find label outliers in your dataset

With Encord Active, you can quickly find image outliers for pre-defined metrics, custom metrics, and label classes. Encord Active finds outliers using precomputed Interquartile ranges.

Prerequisites: Dataset & Labels

Setup​

If you haven't installed Encord Active, visit installation. In this workflow we will be using the COCO validation dataset.

Steps​

Navigate to the Label Quality > Summary tab. Here each metric will be presented as an expandable panes.

You can click on a metric to get a deeper insight into moderate outliers and severe outliers. Severe outliers are presented first in the pane.

Next, you can use the slider to navigate your data from most severe outlier to least severe.

When you have identified outliers of interest use the tagging or bulk tagging feature to select a group of images. After creating a tagged image group, you can access it at the bottom of the left sidebar in the Actions tab.

Within the Actions tab, click Filter data frame on and select tags. Next, choose the tags you would like to export, relabel, augment, review, or delete from your dataset.