Giters

taesungp / contrastive-unpaired-translation

Contrastive unpaired image-to-image translation, faster and lighter training than cyclegan (ECCV 2020, in PyTorch)

Home Page:https://taesung.me/ContrastiveUnpairedTranslation/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

MLP network

CasperDL2805 opened this issue · comments

commented

Hi! First of all, awesome paper. I was wondering what the purpose of passing the feature maps through the MLP is?

commented

The feature map themselves are not designed to satisfy the contrastive loss, because they are just features useful for image translation. But they should be easily projected to another space where the contrastive loss can be met, and the MLP is takes care of this part.

commented

Hi again - thank you for your answer. Can you elaborate on what you mean by "where the contrastive loss can be met"? Thank you in advance.