image upscaling with generative adversarial network

[feisty2]

I recently found an extremely high quality (high quality enough to fool my eyes and make me think the results are native hi-res pics) image upscaling algorithm based on generative adversarial network,,,
paper: https://arxiv.org/pdf/1609.04802.pdf
any one of you wanna implement this thing?

[luquinhas0021]

(high quality enough to fool my eyes and make me think the results are native hi-res pics)
It's not so astounding like you say: a lot of SRGAN-VGG54 results lose a little some contrast and sharpness when put it side by side with high-res images.
Anyway, it's a pretty good image upscaler.
This paper isn't new, I guess, because I've seen it before. The it developer team must be reviewed it.

[feisty2]

and yet it beats the shit out of any other "state of the art" methods, it's 4x upscaling and you don't expect it to be a precise restoration, and by "high quality enough to fool my eyes", I was saying the reconstructed images alone without referring to the ground truth, look like native hi-res pics to me

[luquinhas0021]

and yet it beats the shit out of any other "state of the art" methods, it's 4x upscaling and you don't expect it to be a precise restoration

Yes, I have forgotten the fact of to be a 4x upscaling. Sorry. If was a 2x upscaling, sure it would be a precise restoration.
One doubt I have is if this 4x upscaling was done once or if it was done in two 2x upscalings.

[feisty2]

If was a 2x upscaling, sure it would be a precise restoration
It will never be a "precise" precise restoration, 2x or 4x, as you can only "guess" the lost information, but you cannot literally "restore" it, a 2x guess will definitely be more precise than a 4x guess tho.
One doubt I have is if this 4x upscaling was done once or if it was done in two 2x upscalings.
it was a direct map of 4x scaling

[luquinhas0021]

I know there's no upscaling that recovers the original image. The "precise" was metaphorical.
If the 4x upscaling was done in two 2x upscalings, would the result be more precise?

[StainlessS]

Feisty,
You dont wanna give a crack at this yourself Feisty2, best person on-site for above is in my humble opinion, YOU.
You do seem to have some idea what you are doing (taking the piss a bit there), and not unfamiliar with a compiler either,
best possible person in present circumstance is again, you. If anybody can, you can, (Really).
[EDIT: Of course we wanna an Avisynth version, feel free to also cater for Vaporsynth et al too].

any one of you wanna implement this thing?
There is only one current on-site person who could, guess who ???