4.1 How do I prepare the Ground-Truth data?

Ground-Truth Data is the foundation for an AI model validation. With the datasets defined and identified as "Ground-Truth", we could verify the performance of the specific AI model without hesitation.


The Ground-Truth Dataset(s) contains the "raw data", also the relative labeling to identify the "Truth" definition. For example, on tabular data, a field to describe the category it should be belong to; or on medical image data, a contour to identify the characteristic observed.

The way to describe "Ground-Truth" should be unified on the same Federated Validation project. PI should synchronizes with all Co-PIs, and a role acted as Truther might be required to confirm that all data are "Ground-Truth". Taiwan AI Labs also provides a judging tool to help "truthing" the datasets. Please contact Taiwan AI Labs for further information.