How to generate GT from 300W-LP

[zero0kiriyu]

Thank you for your great work! I am trying to reimplement this repository, but I noticed that you only provide processed training data. I would like to improve the performance on close eye data, so I need to know how to generate ground truth data for training. I tried to infer the preprocessing steps from 300W-LP and the labels you provided, but I couldn’t get it to work. Can you provide the detail how to generate the GT? Thank you!