Yijun Li

i see. Thx for the reply.

Hi @salopge, thanks for your interests. You can collect several thousand of content images as the content is relatively easier to reconstruct (2 pooling layers only). Any image can be...

@suke27 Because I set `diversity_weight=0` by default in `multi_style_transfer_train.lua` :-)

@suke27 Generally it will make the training slower, especially when learning a large number of styles. That's why I set it as 0. I suggest you use a smaller diversity...

@suke27 It's just a random number, you can monitor the loss/results and terminate anytime.

This is not my work. Better to ask the authors :)

--gpu -1

Emm...not sure about your case. I just test it on my PC and the result is good. My CPU: Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz

Not familiar with Alibaba Cloud. Can you try it on a local regular desktop (with Ubuntu)? You may use a small image size if the CPU memory is limited. Can...

This is my result:  Check [here](https://github.com/jcjohnson/neural-style/issues/149) and see if reinstalling the `nn` package will help.