Quality metrics evaluate the quality of various components in your computer vision infrastructure, and therefore constitute the foundation of Encord Active. They are additional parametrizations added onto your data, labels, and models; they are ways of indexing your data, labels, and models in semantically interesting and relevant ways.
Encord Active (EA) is designed to compute, store, inspect, manipulate, and utilize quality metrics for a wide array of functionality. It hosts a library of these quality metrics, and importantly allows you to customize by writing your own “Quality Metrics” to calculate/compute QMs across your dataset.
We have split the metrics into three main categories:
Data Quality Metrics: For analysing and working with your image, sequence or video data. These metrics operate on images or individual video frames and are heuristic in the sense that they depend on the image content without labels.
- Example metrics: Area, Brightness, Blur, Green value.
Label Quality Metrics: For analysing and working with your labels. These metrics operate on the geometries of objects, like bounding boxes, polygons, segmentations and polylines, and the heuristics of classifications.
- Example metrics: Object Aspect Ratio, Occlusion, Object Count.
Model Quality Metrics: For analysing and working with your image and labels with an imported machine learning model. These metrics operate in various ways, some are based on model predictions and other on active learning acquisition functions.
- Example metrics: Entropy, Smallest Margin, Least Confident.
Updated 19 days ago