A random idea that I have a while ago but don't know how to make it: a ML app that can detect if the input image is porn or not.

There are thousands of more or less successful attempts out there for this already.

Example: "safe browsing" in Google

Seems like someone is trying to prepare for Apple interview.

How do you define porn?

@iiii that’s the magic of ML 🙂

@Lensflare that's not. For it to work at all, you need a definition of what's porn in the teaching data. So, how is it defined?

@iiii At first, I use 'NSFW', but I think it is too broad so I choose 'porn' instead. In here 'porn' simply means hentai, ahegao, or any Japanese porn stuff.

Already exists, It's called NudeNet, done in Python. Has pretty good accuracy.

Not perfect, but in my experience It's pretty good when fine-tuned properly

Does it filter out porn, or the other way around?

@iiii not sure if I understand you correctly but the point of ML is to learn the definition by itself by processing training data which is labeled.
So it gets thousands of pictures labeled with porn and thousands which are not porn. And the ML application learns to correctly classify pictures as porn or not porn if you give it some unlabeled image.

So, at no point the definition of porn is given to the ML application, it just learns it “magically”.

@Lensflare you define what porn is the moment you label a training image as porn, that's what he means

@hardCoding well, yes. Obviously 😄

@hardCoding yes, exactly. So you need some definition of what porn is in the first place