Deep Learning

Deep learning thread?
Discuss the best use for deep learning. Imho generating new pepes is the best use of deep learning. Pic related.

Lets see who can make the closest pepes

makerarepepes.me

Other urls found in this thread:

bennycheung.github.io/deep-learning-on-windows-10
github.com/lltcggie/waifu2x-caffe/releases
github.com/lltcggie/waifu2x-caffe/releases/download/1.1.8.3/waifu2x-caffe.zip
waifu2x.udp.jp
affinelayer.com/pixsrv/index.html
junyanz.github.io/CycleGAN/
twitter.com/NSFWRedditVideo

Beautiful

So Deep learning is literally a meme.

...

deep learning is sensationalised by the media because they don't understand it. its basically pattern recognition with trial and error

Thing is really smart!

I tried to draw wojak and it came out horrific.

You don't understand it. It's basically the computational statistical inference.

suck
funny

I tried so fucking hard to install Neural-Style on Windows like a year ago, and I eventually just gave up. I have an Nvidia GPU with CUDA.

it was fucking impossible to set up all these python dependencies like NumPy and Caffe, even after finding Windows specific versions of them.

I just noticed someone made a new version specifically for windows 10 so maybe I will retry it:
bennycheung.github.io/deep-learning-on-windows-10

I fucking hate dual booting so I'm not installing linux over this

garbage t b h

Nice Jew nazi

Looks like it always tries to put the eyes down.

neat

this is what fuels nightmares

...

...

What are some good uses for this? I'm working a little project to boost my portfolio for getting an internship next spring.

>tfw no framework works on a fucking and GPU

Its annoying really.

upscale it, paint over it a bit with acrylic so its not noticeable

sell modern pepe art

Guys check this shit out. You can magnify images if you have an nvidia GPU and it interpolates all the missing detail when it blows the image up.

Windows version:
github.com/lltcggie/waifu2x-caffe/releases
Direct link to download:
github.com/lltcggie/waifu2x-caffe/releases/download/1.1.8.3/waifu2x-caffe.zip

It's pretty neat. This picture from Calvin and Hobbes was blown up to twice its original size (maybe even 2.4x - I made it a while ago) and it only took like 2 seconds on my old GTX 680

...

>tfw fat pepe man

How will this help me get a CS internship? Perhaps I can send nice gift baskets to companies with obscure pepe murals...

Also I meant deep learning, not the pepe generator itself lol

...

oh yeah this wont help you at all then

generate landscape concepts on the fly

>What are some good uses for this?
Judging by the images ITT, they could use it to generate the monsters in the next Silent Hill game.

Like, of paintings? I'm thinking about making this but instead of pepe, you out in a term. It then scrubs Google for the first however many thousand images from searching that term and generates something, perhaps also based off of drawn input.

...

wew lad

looks like pepe overlaid onto jew hands desu, nice job

...

honestly, whatever I draw is no better than a random seed right now

...

I'm not going to the website on the off chance this is some elaborate ruse to give my computer aids, but I'm surprised nobody has thrown dickbutt at it yet

what is a dickbutt haha?

Someone should make a version of this that magnifies every frame of an animated GIF and then lets you render it together as a webm

a doodle young people draw.

>what is a dickbutt
You're rusing me right?

pls kill me

first off, thank you.
ive been looking for some shit that does exactly this as well as this does it

a reddit/imgur meme

...

...

>Sup Forums only now learns about waifu2x

Was this just an Sup Forums thing? This shit has been around for ages.

>reddit
fuck you dickbutt is a Sup Forums treasure

it looks random af

Did the same thing.

whatever man, but like ive been looking for a way to upscale all those 720p scans of ralph mcquarrie concepts

which, good luck finding that shit not in 720p

Hlpp me

>if you have an nvidia GPU
My bad, you can use it with just your CPU as well. It just takes a little longer per-image than if you set it up to work with CUDA

I first heard about it here on Sup Forums a little under a year ago. I'm sure you're aware, but there is also website that does it all for you in-browser:
waifu2x.udp.jp

It doesn't have as many options as the downloadable version though

...

mami's mammaries

too easy

its shit

i tried

I wanna train a NN on pics of Miami!
Infinite Mami!

These almost look like female torsos.

He looks sad

HE SEES ALL THAT YOU HAVE CREATED, MORTALS, AND HE IS NOT IMPRESSED

>waifu2x.udp.jp

...

...

...

...

not anymore fuck boy

garbage

fuck you man. Dickbutt will never belong to Reddit. Neither will advice animals

>makerarepepes.me
This website is trash. Nothing comes out looking anything like pepe

show me something cool

...

...

Gettting lots of twisted looking heads.

Random, I think not

curly vertical lines sometimes create geniune nightmare fuel

here's another website that does the same thing but better, except instead of making pepes it makes buildings, shoes, cats and handbags:
affinelayer.com/pixsrv/index.html

yes

a slightly different input results in an output that is also only slightly different
adding a little bit of garbage lines to your input and putting consecutive outputs into a gif yields some fun animations

neat

scheisse

Would be nice to have undo or an eraser tool or something. After a couple of frames everything becomes a mess and the output is nothing more than a green blob with artifacts. Still kinda entertaining.

I'm pretty sure that FaceApp is powered by either DiscoGAN or CycleGAN
junyanz.github.io/CycleGAN/

is there no online implementation of faceapp? Or anything similar even?

Wow.

Du bist scheise

FUCKING IDIOT YOU MADE A SIMPLE FUNDEMENTAL ERROR

this would work way better if you fixed your edge detection on the input ffs

of course, you didn't make this, you just copied "edges2cats" except not exactly which would be a good idea but you fucked up something very simple with it showing how dumb you are you stupid idiot retard

these are some abstract pepes

Helly pepe.

...

hmm..

...

Huh, you can see some similarities, I reckon it's the training data.

...

...

wow, it sucks

give some more information user. whats the network? i knew that many networks for recognition tasks have some sort of cnn followed by a rnn and the classifier.
is this the reverse? penpoint->rnn->cnn->image?

What happens if you use pepe as an input?

You couldn't make one that actually looks like pepe if you tried

>Can't turn back time

...

i tried