Project Score

Project Score Definition

The “Project Score” is a universal metric that offers a quick overview of the status of a project and its models. Specifically, it is computed as the performance of the active model on the test subset; in multi-task environments (“chained tasks” in the Intel Geti platform), the score is obtained by aggregating the active models of each task.

Metric Specificity per Task Type

Different task types require distinct evaluation metrics due to the unique nature of each task:

• Classification: The metric of choice is “Accuracy”, which measures the proportion of correct predictions made by the model.

• Detection: “F-measure” (also known as F1-score) is used, combining precision and recall to provide a balanced view of model performance.

• Segmentation: The “Dice Score” is employed, assessing the overlap between predicted and actual objects to gauge accuracy.

Usage Considerations

Given the variability in evaluation metrics, the “Project Score” should be interpreted as an indicative rather than an absolute measure. This approach ensures that the score not only provides a snapshot of model performance but also acknowledges the varying scales and implications associated with different metric outcomes.