After receiving dataset. I figure out that train and valid only contain image without labels.
The goal of this challenge is to generate realistic images from semantic segmented images.
So please, anyone can clarify, what is the role of train and validation dataset in this case?
Please check the FYI regarding your question. Let us know if you have other questions.