Modalities Handbook#
Introduction to modalities#
When you create a datapoint on Datagen’s platform, that datapoint takes the form of a rendered image of a synthetic human face. But to train your model, you need to compare that image to an underlying ground truth.
Because the images are computer-generated, we are able to provide perfect ground truth data in separate files that you download along with the images. Each file annotates the image in a different way, to bring a different part of the ground truth to the forefront: the locations of facial landmarks; a normal map that reconstructs the contours of the face; the age, gender, and ethnicity of the subject; and so on.
Each set of annotations is called a modality: a method of looking at the data to highlight a different part of the ground truth. This document details the structure and format of each of these modalities, so you can process the data properly and import it into your training set.
Table of Contents#
About the modalities
Visual modalities
Keypoint modalities
Other modalities