AI upscaling using a sample as reference?

[CGI979]

I have a 5 minute clip which I had converted to 320x240 a while back. Fast forward, where I was editing the source (1080p) and accidentally overwrote the the original with edit.

Now I have the following:
- Unedited full length clip in low resolution (320x240).
- Edited partial length (60% of the original) clip in full HD (1080p).

Question is, is there a free program I can use the HD version as reference and upscale the full length clip? I looked into Waifu2x, however I would need an nvidia video card, which I do not want to waste money on yet unless I case see if the quality will be worth it.

Any recommendations? Yes, I tried running some free recovery tools to recover the original, looks like its gone.

[manono]

You don't need an Nvidia card to use it:

http://avisynth.nl/index.php/Waifu2x#Resizers

but I don't think using it will help you all that much. From 240p to 1080p?

[poisondeathray]

waifu2x is limited to "2x"

Some of the github deep learning research projects, such as ESRGAN can train 4x models

https://github.com/xinntao/ESRGAN
https://github.com/yulunzhang/RCAN
https://github.com/thstkdgus35/EDSR-PyTorch
http://mmlab.ie.cuhk.edu.hk/projects/SRCNN.html
https://github.com/Zheng222/PPON

many more...

But it can take a long time to custom train the models. Some projects have pre-trained "generic" models that you might want to play with to test it out



Single image based algorithms tend to have temporal consistency issues, flicker. TecoGAN is supposed to help with that for video, but I couldn't get this to run
https://github.com/thunil/TecoGAN