DeepCreamPy AI is our savior

Someone took the DeepDreams AI and made it so it uncensures images
here is the github link if you wanna check it out
github.com/deeppomf/DeepCreamPy

I will just come clean and say I saw this is in leddit and Im surprised no one is talking about it here

Attached: DeepCreamPy.png (1830x1470, 1.18M)

Other urls found in this thread:

arxiv.org/abs/1804.07723
github.com/ROCmSoftwarePlatform/tensorflow-upstream
twitter.com/NSFWRedditGif

There was a thread 2 days ago or so. I think the OP claimed to be the author, even.

You need to go back, we already had threads about it newfag

Oh ok then didnt know about it, dont come here on Jow Forums often

Then why the fuck do you use the exact same image every time. We know it exists you have already succeeded in shilling it.

lol wtf everytime? first time I post this and the image is an exemple of the github page

We have threads in /h/ and /d/ right now and there was a thread in /a/ and there was one here on Jow Forums a few days ago.

>train inpainting network
>make millions

I tried it, it's fucking broken. Just won't run without TensorFlow spitting a fuckton of dll errors at me. And yes, i'm using Python 3.6.7

Pretty sure tensorflow requires CUDA 9 and cuDNN...

Shit, i guess i'll have to wait til my psu gets back from repair

How does 'training' for this thing work, anyways?

yeah im curius as well but I think is something to do with pattern recognition and im prob not smart enough to understand

>CUDA 9
they change the build requirements all the time, and those are only the prebuilt sources, you can compile it with your own hardware if you really wanted to

ML models are unironically statistical models.
Take a look at every "linear regression" example for machine learning.

They start with linear regression because that's all ML is.
You take a bounded shape formed by a complex function and try to fit it to your data.
Sometimes your function isn't complex enough to fit your data, so you make it more complex or you tinker with it a bit in a way you think makes it fit better.

Try "CUDA_VISIBLE_DEVICES= python decensor.py [whatever]" instead to force it to use CPU

arxiv.org/abs/1804.07723

Can people stop using proprietary libraries like CUDA and start using openCL or something?

t. amd employee
go beg on the tensorflow github issue tracker

would you look at that
github.com/ROCmSoftwarePlatform/tensorflow-upstream

Have you ever done linear algebra? ML is basically a fancy least squares solution.

Go home redditor.
We've already seen this, enjoy your regurgitated Jow Forums content on reddit instead you unbearable pleb.

its not even good

Just use the photoshop repair tool

for all intents and purposes this is just the spot healing tool in ps, but somehow worse in every way possible

OpenCL doesn't support NVIDIA cards on IBM power systems
You don't want to exclude IBM customers, do you user?

>Irregular Holes
I guess you could call them that...