Giters

HsinYingLee / DRIT

Learning diverse image-to-image translation from unpaired data

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Different batch size

XuYunqiu opened this issue · comments

commented

Hi, thanks for your great work about GAN.
I found in your paper you said you use batch size of 1, but in this code the default training batch size is set to 2.
I would like to ask is there any special training trick about batch size?

commented

Hi,

Please refer to model.py, we use self.real_A/B_encoded as the input images for the main training process (disentangled representation, cross-cycle consistency). As for self.real_A/B_random image, we only use it to train the discriminators in latent regression (refer here). We found that this would yield more diverse results.