Taking requests.
A.I/Neural Network Picture Upscaling
Other urls found in this thread:
github.com
youtube.com
waifu2x.booru.pics
u.lewd.se
twitter.com
please upscale this image kudasai
one 6x6 pixel for u bub
Upscale this text pweasee.
Link me the github page
>A.I/Neural Network
stop this meme please
next thing you're gonna say is that fizzbuzz that draws a line in photoshop is a decensor AI
left is conventional resizing
I wasn't even referring to your fizzbuzz, but your fizzbuzz can be achieved by resizing and using anti-noise effects
"A.I" is literally nothing more than a spreadsheet of if else
Now you and all other autists that wrote their first python script or managed their first databab, fuck off and go shill somewhere else, preferably with more buzzwords
elaborate.
What he's trying to say is, post code
I wanna know that the algorith scans the pics, colors and its positions, and redraws the picture in quality resolution
Thats AI
What's not AI is the script resizing/denoising
He's using waifu2x, mate. Even linked it in this thread already.
ai dick enlargment.
...
Exposed yourself as having no experience or understanding at all about how neural networks work. NNs and conventional ai are two completely different paradigms.
try this
thanks
r
this next
My dick's bigger.
make it big ai man
Kindof interested to see how well it works on a busy image with lots of text..
AI Size comparison fetish.
hard mode
Not even hard. The only issue is the picture is innately noisy. If that image were devoid of that noise, even at that size, you'd see better results.
is it possible to upscale and denoise videos by altering every frame?
Very possible!
Yes, you can finally watch the original naruto in 1080p
impossible mode
give me half an hour
what should I look for?
I'd like to get a better picture quality of an old movie I got, even the original dvd release has terrible capture resolution poor focus and washed contrast
Are you gonna process each frame as a separate image then stitch them together?
you FUCKING BET.
You're talking around ~30,000 frames, assuming a movie is only an hour long. You'd have to upscale each individual frame and them render them in sequence again. Partially not viable, but I'm sure there'll be innovations in that regard in the near future.
Check out Nvidia's frame interpolation: youtube.com
There should be some convenient script to do it automatically.
Folder with all the images of a movie + script using the denoising tool + time = folder with bigger images and you just compile it together.
Requests on hold for a bit....
>creates oversized versions
>can't trivially tell which is original when reverse searching
REEEEEEEEEEEEE
How long does it take per image to run?
I'd bet you could write a script to do it, and then just let your computer chug for however many days.
Around a minute, give or take. All these images are going to be scaled up by 600%, so it takes a hot second.
Now, if we're talking 30,000 frames, (assuming the movie is only an hour long), that's 20 days my guy.
welp is just an experiment I guess I'll do some research, thank you
nn's are literally the same shit
so what, you "train" your pajeet AI with pictures, much skill
If we took a movie that was initially 960x540 and only doubled you're talking probably half or more reduction in time to process to make it 200% bigger (1920x1080).
The ability to take an image and scale it as you wish to whatever whilst preserving its quality near ground-proof is one hell of a fucking skill.
You'd have had to commission this kind of a thing to an artist if you wanted to do anything like this at all. Expensive, time consuming, labor intensive, and inefficient.
This depends on your specs though.
Something like this is highly scalable, so if you had access to a threadripper or something you could really cut that render time down.
Very true -- I was referring to just my capability.
What CPU are you running these on?
Actually using my GPU to process them, not like my CPU would help. It's really bad, in terms of wanting to actually do stuff intensive with it. I've a Titan X, however it's reaaaally old and I'm way past due for an upgrade.
>>requests
But anyone can just do it online:
waifu2x.booru.pics
I'm using my own trained model. Not to mention the model that they trained waifu2x on is intended for anime/manga stuff, not actual photos.
Well shit, if a Titan X takes a while to make these, scaling up would be expensive.
Actually, thanks to Nvida, it's a lot less expensive already. They've just release their Cloud Computing suites which use things like their just announced T4's, which all use the new Tensor cores. I believe they've sped up overall data processing by 6 times? Something pretty insane like that.
seems to fall apart here
Actually, its still doing its job. The issue is that the source material has noise, which gets replicated once the image is up scaled.
you will notice that often straight lines get over-corrected and seem to worbble a bit, which I've noticed with certain images.
Close to finishing this one.
I'm genuinely surprised it came out as good as it did -- usually it muddies pixel art pretty bad.
Only 3 frames, shouldn't be hard to make it postable.
on it
Excellent, the sataniaposting collection grows.
adjusted the speed a bit
I specifically gave you a pixel art gif cuz I knew it wouldn't turn out good, but you're right; it turned out pretty good.
dab
This is it chief