Dataset structure and formating

[ArpanGyawali]

Hi,
I am working on LVO detection from 2d MIP images,
I went through the doumentation but didnt find a way, how to strcture a custom dataset. How are training and testing set arranged and what type of label formatting it expects?

The documentation suggest to apply preprocessing script to convert the data to trainable format, but first of all how the data and annotation labels is structured. Please can someone help.