Daggerfall Workshop Forums

In my retexture thread, a guy named Poberun dropped by and showed this very impressive result:
viewtopic.php?f=14&t=632&start=20

After I asked him how he did it, he revealed that he used two programs to do this:
"I used https://github.com/nagadomi/waifu2x for upscaling original textures, then I used https://github.com/alexjc/neural-doodle with some googled stone-handprinted texture to create underlayer, lighten it up a bit, after that I overlayed upscaled texture over it and finally used waifu2x to upscale it once more"

Now, I asked in that thread and I wrote him a PM but no reply.

What I NEED to know is, HOW to use those programs???

They dont have any GUI, any *.exe, nothing... and the documentation has not helped at all.

Can someone shed some light on this topic?

I use windows 7 x64 and nvidia gfx card

Thanks in advance!

Looks like they're both Linux based. That said, I found this post in the "Issues" tab for waifu2x:
https://github.com/nagadomi/waifu2x/issues/202

Which points to a Windows version here (includes English UI):
https://github.com/lltcggie/waifu2x-caffe/releases

Program details here:
https://github.com/nagadomi/waifu2x/issues/201

No idea if you can get neural-doodle working in Windows.
If you had Windows 10, it supports running Ubuntu natively (not a virtual machine).

I never used them but there are also a few websites (like this and this) which claim to implement this algorithm.

I haven't used these tools, but from my experience using neural network libraries with CUDA for other purposes, it may be easier to install Ubuntu in dual-boot in your machine* and following the instructions than to try to make them work on Windows.

Libraries like Torch or TensorFlow do run on Windows, but Windows support is always an afterthought, you get better documentation and community support if you use Linux. I am mostly a Windows user myself but I tend to boot into Linux or connect to a Linux computer when I want to do stuff in the GPU with CUDA. I once tried to install TensorFlow on Windows but I just wasn't able to make it run on GPU. Perhaps Torch is easier though.

With Linux installed, the documentation for these programs looks quite good. The first program actually tells you how to install all dependencies, etc.

*The Ubuntu subsystem for Windows does not have CUDA support. You need a full-fledged Linux install to do this kind of GPU computing. See https://devtalk.nvidia.com/default/topi ... /?offset=6 and https://wpdev.uservoice.com/forums/2669 ... pu-support

Thanks for input guys! Im getting some results now, posted it here viewtopic.php?f=14&t=817&p=9560#p9560