Updating

Updating a deep learning model allows new classes to be added without retraining on the entire set that had been used up to that point.

However, our testing has shown that an extensively trained and validated model is required for updating to work properly.

Whether adding new classes or not, unless the model has been trained on at least thousands of previous images, we recommend to refine models by adding the new tracings to the original Session, and training a new model on all images using Train New in the AI Session Processor.

*Tip: When updating, make sure that your new tracings contain some features that look like those from the old images, otherwise the model may start forgetting them in favor of the new ones.