You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
After spending a lot of time and frustration I finally found the problem why bad results are given. The problems lies in the whitening and color transformation function more specifically the way the style feature map means are added back to the colored feature map at the very end. The implementation is different in this repo and in PytorchWCT repo. Although it's been quite some time since the issue was raised hope it helps
For content
![content](https://user-images.githubusercontent.com/10587992/54868006-bf380c80-4d98-11e9-80ab-9bda0d8b00d5.jpg)
and style
![wave_crop](https://user-images.githubusercontent.com/10587992/54868011-ce1ebf00-4d98-11e9-9e1c-5cb797d31bf8.jpg)
the method gives bad quality result:
![000000001584_stylized_by_wave_crop_alpha_1 00](https://user-images.githubusercontent.com/10587992/54868016-df67cb80-4d98-11e9-9c46-ca1077c4318a.jpg)
which look like a bug and not expected behaviour.
See result by https://github.com/sunshineatnoon/PytorchWCT:
![1_1 0](https://user-images.githubusercontent.com/10587992/54868043-31105600-4d99-11e9-8762-04d84f9623fe.jpg)
I used the command:
python main.py --content content.jpg --style style.jpg --outDir results/ --alpha 1.0
.The text was updated successfully, but these errors were encountered: