Artist here. I've got a question about this project

[jcklpe]

Hi!

I'm an artist and designer (and a sorta crappy dev) who fiddles around with Deep Style in my spare time.

I recently have been doing a bunch of articles focused on teaching other artists how to install JCJohnson's neural style. Someone commented on one of my facebook posts saying I should check out nvidia dockers. I didn't know you could use GPU with docker now!

So my question is... is that what this is? I see a bunch of people online writing neural style dockers for AWS but I want to just run the docker on my windows machine. Is that possible and still have good resolution etc?

[albarji]

Hi there! Happy to know an artist is interested in this project! I'm just a dev and a sorta crappy artist after all. :)

So answering your questions, this project does use nvidia-docker to easen the install, so you don't need to go through the burden of installing all dependencies. Just docker, nvidia-docker and the relevant drivers for your GPU (check out the install instructions).

Two warnings though. First, I have never tested this on Windows, and I can't find any reports on successfully using nvidia-docker on Windows. So, you should still go for Linux. And second, the resolution of the pictures largely depends on the amount of memory your GPU has. Even if you have a small GPU this project will make an effort to split the picture into smaller tiles, render each of them separately, and them stitch them back together. Sometimes this will work nicely, sometimes the "seams" will stand out. But I encourage you to try and see how it works for you!

[jcklpe]

Excellent, thank you for the response! I'm aware of the patching version of neural style. Is that the only way the docker can be used? I prefer to use the smaller images and then blow them up using something like Waifu 2X Sincerely,David Aslan French Jackalope.tech Design X Code: Pushing the boundaries of design, art, and code. jackalope.tech Find me on Facebook  Let's connect!  On Sun, May 20, 2018 5:18 AM, Álvaro Barbero Jiménez notifications@github.com wrote: Hi there! Happy to know an artist is interested in this project! I'm just a dev and a sorta crappy artist after all. :) So answering your questions, this project does use nvidia-docker to easen the install, so you don't need to go through the burden of installing all dependencies. Just docker, nvidia-docker and the relevant drivers for your GPU (check out the install instructions). Two warnings though. First, I have never tested this on Windows, and I can't find any reports on successfully using nvidia-docker on Windows. So, you should still go for Linux. And second, the resolution of the pictures largely depends on the amount of memory your GPU has. Even if you have a small GPU this project will make an effort to split the picture into smaller tiles, render each of them separately, and them stitch them back together. Sometimes this will work nicely, sometimes the "seams" will stand out. But I encourage you to try and see how it works for you! — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[albarji]

Hi there again,

You can control the size of the image produced by this neural-style implementation. For small images I don't make use of tiling, so you can get a single neural transfer and then apply other tools to increase the image size. How big an image can be to fit into a single render depends on your GPU model and the particular style transfer algorithm you choose. Here are some values for some common GPU models, assuming square images: https://github.com/albarji/neural-style-docker/blob/master/gpuconfig.json