Soup Jow Forums,
I am the proprietor of an online forum similar to this one and am interested in finding a way to keep CP and other trash off my website.
I was thinking of scraping both porn and work-safe images off the internet and using that to train a neural network to recognize non work-safe images, which should be fairly easy to do.
Any advice, tips or suggestions?
Other urls found in this thread:
en.wikipedia.org
microsoft.com
twitter.com
OP again, I'm no expert in machine learning, and neural networks seems like some magic 'one size fits all' solution to me, and this seems like a quick and easy way to do it, even if it's not the most efficient. Thoughts?
Yeah dude training a neural network with cheese pizza to moderate your site seems like a great idea.
Lmao, was about to say the same thing
>scrapping porn image off the internet
You'll need child porn, which is illegal.
the only way to train it to block child porn is to actually download and give it child porn. Don't do that
There's definitely ways to do it without possessing actual CP
Feeding it pics of children and pics of nude adults, then checking if a certain pic matches with both?
Problem statement too complex. NSFW means everything from gore to porn. Make several classifiers instead.
I wish someone would feed me pics of children and nude adults
The way NNs end up processing images to recognize them is exceptionally bizarre and unintuitive. It's highly possible that even if it can detect legal porn 99% of the time, it will fail to detect CP at all unless you train it on CP. There's probably ways to do it without training with CP, but not for a novice importing a neural net framework into Python. Consider that it's very likely that companies that developed NNs to recognize CP were contracted/sponsored by the government to do so and were given access to CP volumes to train with.
Good luck getting the training data.
Couldn't you use that nvidia "these people aren't real" generator/GAN thing to generate artifical cheese pizza?
I think it would have to study their nude bodies first in large quantities, so not really.
Or you could use computer animation to generate CP and feed it into the NN. This is probably illegal though
Yeah, just downloaded 1TB of CP first. Just tell the feds you're using it as a data set, they're cool with it.
Apparently access is not freely given, you have to apply and if you're not a big service or corporation they don't give it to ya.
Install gentoo
Try using a pretrained model like open_nsfw.
Depends on where he is, really. But really it wouldn't be hard to get the training data from Tor. But if he's caught with it... The bigger problem is self-hosting a neural network powerful to do sophisticated image recognition like that.
PhotoDNA is also not AI, and relies on images having been previously discovered and added to a database. It doesn't catch OC.
Basically this. You won't be able to train it yourself anyway
