search

hwalsuklee / tensorflow-fast-style-transfer

A simple, concise tensorflow implementation of fast style transfer
Apache License 2.0
245 stars 70 forks source link

Quality compared to style-transfer #5

Open exsstas opened 7 years ago

exsstas commented 7 years ago

Hey!

I'm trying to achieve with fast style transfer roughly same results as with classical one and I can't. Best example, IMO, is style transfer from "Starry night": in Gatys approach (your implementation is fine example) you can see spirals on images styled with Van Gogh style. With fast style transfer - there is no spirals, just some boring zigzags.

I upload trained model with default parameters here. Before that I worked with Logan Engstrom's variation of it and did experiments with different content weights and style weights, but cannot achieve spirals transfer as well. In your code there is possibility to play with layers as well, but it's time consuming to try all possible options. Can you suggest what should I try to achieve result closer to style transfer?

PS: It's not about only achieving spirals in "Stary night" - it's just a very good illustration of an issue.

hwalsuklee commented 7 years ago

Hi. I'm really sorry for late response.

I have been faced with the same problems that you mentioned. Honestly speaking, I could not achieved spirals in "Starry night" from the codes here.

I did not want to spend too much time on achieving "good" results as in other implementations. So I left it as further work.

About layer-option, I gave some variations on layers just for checking that codes are working.

I'm sorry I can not help you.

nlminh commented 7 years ago

Hi, with your pre-trained models: hier which values for parameters did you use to train? Thank you.

hwalsuklee commented 7 years ago

Hi @nlminh All pre-trained models are obtained from the same default parameters.

scharnk commented 5 years ago

Hi, I'm having trouble recreating the same results using the default parameters, has anyone else experienced this? I'm wondering if having content loss = [1.0] is too much?

dakshvar22 commented 4 years ago

Hi @scharnk, Yes default params seem to be off for some reason. Have you been able to reproduce the pre-trained models?

SolicTous commented 4 years ago

@dakshvar22 @scharnk, what about reproduce models? Can you say - what i must to change?

@exsstas Agree with you. I'm interesting how to reproduse demo results from first version of @hwalsuklee on this repo. Because resuls is really differ.