SRGAN-tensorflow icon indicating copy to clipboard operation
SRGAN-tensorflow copied to clipboard

Published 20 hours ago • verified publisher icon brade31919

Data Normalization

Open YaanGe opened this issue 6 years ago • 2 comments

Why do you normalize the low-resolution image to [0, 1] and the high-resolution image to [-1, 1]

While feeding data into VGG network, should it be normalized to [0, 1]?

YaanGe avatar May 31 '18 13:05 YaanGe

Hi, I have the same problem with you. And have you figured it out?

seemaywang avatar Nov 30 '18 03:11 seemaywang

I didn't follow author's code at this part, instead I normalized both low resolution and high resolution images to [-1, 1], and normalized the data to [0, 1] for VGG loss computation. The result is good with my dataset.

YaanGe avatar Dec 03 '18 05:12 YaanGe