Skip to content

Input preprocessing

Define your preprocessing pipeline once in Python, train on its output, and pass it to dg.export. The exact same steps compile into the model graph and run on-device, with no divergence between training and deployed preprocessing.