fast-style-transfer icon indicating copy to clipboard operation
fast-style-transfer copied to clipboard

Style.py: Question about Python version and operating system

Open beinando opened this issue 4 years ago • 1 comments

Hello,

i just try to get this style.py thing running, and it trains without any error message. Unfortunately, the outputs look wrong and it seems that the algorithm is converging to a weird state.

This is an example images i get after training for around 1 hour on a geforce 1080 TI: (content = chicago.jpg, style = la_muse.jpg)

8_10

I already tried to resize the style and content image to 224x224 pixels but it did not help.

What could be the reason behind this bad results? Do i need a linux system to get this running? Can i use python 3.7 , and should it work for tensorflow 1.15.0 ? If someone has some information for me, would be really cool! Thank you!

beinando avatar Mar 12 '20 14:03 beinando

For content you are recommended to use COCO's train2014 folder which includes more than 80k images instead of just 1. That's why you are getting those results.

apardogi avatar Jul 14 '22 09:07 apardogi