Promoting Training Results as a new "Model"

One common task in model training is to reach a point in your training where you believe a resulting checkpoint is now potentially usable as a new model. However, you can treat this as an iteration on the way to your final model, and you may want to run a new set of training, perhaps with a new configuration, on this checkpoint.

In XNAT, you can do this by "Promoting" the training results as a new Model. You can do this in the Model Training Dashboard.

1. Click "Promote Training" in the Training Results table.

2. Name your Model and Select Which Outputs to Collate. A dialog will open that allows you to name your new model. You can also decide whether to accept the "final" model results, or the "best" model results. You may want to Viewing Training Progress in TensorBoard to determine which would better serve you.

All models must have unique names. Attempting to use an existing model name will return an error.