cysmith / neural-style-tf

TensorFlow (Python API) implementation of Neural Style
GNU General Public License v3.0
3.11k stars 828 forks source link

About Quick Neural style #25

Closed dovanchan closed 7 years ago

dovanchan commented 7 years ago

I have learned many things from your code! but I felt a little slowly when run your net in GPU gtx 1080(about 10 mins for one picture). I heard that there are some network that can implement The fast Neural Style and get the nearly result。Can you introduce some new and great github code for me ??I wanna learn more about the Neural Style!~ Really Thanks for you coding!~~

MarcoForte commented 7 years ago

This might be what you are looking for https://github.com/lengstrom/fast-style-transfer The training takes longer but once you have trained the network you can apply that style to any image and it will be fast

dovanchan commented 7 years ago

https://github.com/tensorflow/magenta/tree/master/magenta/models/image_stylization

This us Google's New projet.How do you think about this? Can you find some another code which can train multiple style to one model and then quick feedforward~

MarcoForte commented 7 years ago

Yes the google project is an extension of the one I linked.

This can transfer arbitrary style https://github.com/rtqichen/style-swap

dovanchan commented 7 years ago

yes,thanks.But I wanna show you a amazing website. https://dreamscopeapp.com/ this style transfer, just take 30s ,and you can upload any style whatever you choose.it still just spend just 30s!!!! and the pic is really beautiful, do you know how they did that?share me the code~~

dovanchan commented 7 years ago

I have tried many neural style transfer code,But I think the photo you generate is really beautiful ;but the only disadvantage is about the speed and it can not save the model for next content ; So,can you recommend another github which can be faster and save model ( almost the same transfer result with you )

dovanchan commented 7 years ago

use gtx1080 ,about 500s to generate a picture.How about you?

cysmith commented 7 years ago

Please keep issues related to problems with this project. There isn't much I can do about the speed of TensorFlow.