visipedia/imat_fashion_comp

label difference between train and val data

JiangYunchao opened this issue · 0 comments

I've noticed that there're only 'color', 'pattern', 'material' have multi-label on train dataset. but all class have multi-label on val dataset. Even 'category' class has only one image which has multi-label. Is this a kind of noise or different data distribution?