Preparing good training material is the key for training a well-performing neural network. How to set-up the training data depends on the use case.

When training a neural network to distinguish different materials or identifying fibers in a GeoDict structure file (*.gdt), there are two possibilities to set up training data.

1. GeoDict can generate training data for material and object identification using its powerful structure generators. For this, a generation script must be created as explained here. Then this script is used in Design of Experiments and Create Training Data to generate the desired amount of training data.
2. If other training data is available from segmented scans the training data must be set-up in a defined folder path, so that GeoDict recognizes it. How to do that is explained here.

If instead a grayscale image should be enhanced, the training material must fulfill certain requirements which are shown here. Save the training data in the defined folder path.

To train a neural network to segment a grayscale image, the data can be generated with GeoDict as shown here. Then, organize the training data as shown here.

In all cases, use the created training data in Train Neural Network.