We refer to user created as positive . We also created a set of neutral outfits for each user by randomly mixing the tops, bottoms and shoes. For each user, we split the user created outfits and neutral outfits into three , i.e. the , validation, and testing sets.

So they tried to turn this into a supervised problem by creating a dataset of “combination created by a person” and “combination selected randomly”. Seems a little rigid.

Source link
thanks you RSS link
( https://www.reddit.com/r//comments/9mynlo/1802443_fashionnet__outfit/)


Please enter your comment!
Please enter your name here