Excuse me ,is there any Caffe implementation code?

[LouiValley]

Excuse me ,is there any Caffe implementation code?

[VaKonS]

@LouiValley, this one? https://github.com/BVLC/caffe

[ProGamerGov]

@LouiValley As per the name, CycleGAN is a generative adversarial network (GAN) style transfer system. Neural-Style is a convolutional neural network (CNN) based style transfer system. CycleGAN is also not capable of making changes to image geometry like Neural-Style is.

A simple search shows this project as the main CycleGAN project: https://github.com/junyanz/CycleGAN

[LouiValley]

Sorry, I may not explain my problems well. I mean is there any code about Neural-Style that can run in Caffe? Cause there is only code can run in tensorflow. what is more , dose Neural-Style use deconv in layer? :） Thanks!

[htoyryla]

"I mean is there any code about Neural-Style that can run in Caffe? Cause there is only code can run in tensorflow. what is more , dose Neural-Style use deconv in layer? "

You are mistaken. Neural-style is implemented in lua using torch; it uses models in caffe format which it reads in using loadcaffe. There are no deconv layers.

I guess by "run in caffe" you mean something like writing python code using pycaffe. I am not aware of any neural-style implementation done that way. When I use python, I prefer to use pytorch.

[VaKonS]

@LouiValley, I guess, you mean something like this, but for "neural style": https://github.com/msracver/Deep-Image-Analogy

There are some neural projects reviews, and as far as I can tell, no one tried to implement it in Caffe yet. https://handong1587.github.io/deep_learning/2015/10/09/fun-with-deep-learning.html https://github.com/ycjing/Neural-Style-Transfer-Papers

[Naruto-Sasuke]

@VaKonS Why not just write the code yourself, It is not difficult.

[fastlater]

@VaKonS Any tensorflow implementation? I found many fast style transfer but I still cannot find one where I could set the weights.

[VaKonS]

@fastlater, here are at least 2 projects of normal style transfer in TensorFlow: by @anishathalye and by @ckmarkoh.

As for fast style transfer, the style and content weights are saved in model during its training, if I'm not mistaken. Here are the options in @jcjohnson's project to train fast style model: https://github.com/jcjohnson/fast-neural-style/blob/813c83441953ead2adb3f65f4cc2d5599d735fa7/train.lua#L35-L45

And this fast style transfer in TensorFlow also sets content and style weights during model's training: https://github.com/lengstrom/fast-style-transfer#training-style-transfer-networks

[VaKonS]

@fastlater, come to think of it, maybe there is a way to make fast style transfer with variable strength: https://github.com/Heumi/Fast_Multi_Style_Transfer-tensorflow

You can make multi-style model with two styles:

• 1st style with full weight;
• and 2nd style with zero weight (--style_loss_weights = 0).

And then set weights for "full" and "zero" styles with "--style_control_weights" parameter. I didn't try this though.

[fastlater]

@VaKonS Thank you for your reply. Very useful links. I will check them all and I am sure one of these will be good for my application.