A dataset must be configured into a dataset folder containing two directories named classes and annotations. The classes directory contains a folder for each classification filled with CSV files of data in each column. Multi-channel classes require multiple data columns. The annotations folder contains three text files instances_test_list.txt, instances_train_list.txt, and instances_val_list.txt. These files are used by model composer to split the dataset into training, testing and validation sets. It is recommended that 50% of the files are in the training list, 30% in the validation, and 20% in the testing list. A visualization of the architecture is shown in Figure 4-5.

Figure 4-5 Folder Path