Live data from Hacker News

Decensoring Hentai with Deep Neural Networks

github.com

61–70 of 73 posts

Re: Decensoring Hentai with Deep Neural Networks

#61
post #23
post #9

Earlier quoted context omitted.

There's tens of thousands of manually decensored manga, you could download and compare them to original to identify censored regions and use as training I suppose.

Is it bad form to share one of these "datasets" here?

Asking for a friend?

Re: Decensoring Hentai with Deep Neural Networks

#62
At least for the most common forms of obfuscation, pixellation and fuzz, why do you need AI for this? It seems conceptually simple to interpolate from several similar but moving frames, how a small obfuscated feature travelled across a huge pixel boundary to cause the small change in the entire oversized pixel.

You'd need I, be it AI or the normal kind, to invent a guess from a single frame. But watch frames that each differ a little over time, and you should be able to deobfuscate pixellation and fuzz with plain math.

Conceptually simple. Not saying I could knock it out over the weekend, just that there doesn't seem to be a mystery about how to proceed that puts it into "throw some ai at it and see if anything happens by ai magic" territory.

Re: Decensoring Hentai with Deep Neural Networks

#63

There should be NSFW examples in the Readme. I saw some practical (NSFW) examples on 4chan and it doesn't seem to work very well. Most hentai doesn't contain simple lines and blobs covering homogenous skin area. There is detail under those areas that the algorithm does not reproduce at all.

On the contrary, the examples I saw on /h/ looked pretty impressive for the most part.

Well, I guess we have different definitions of impressive. The de-censored areas did not reproduce genitalia properly at all.

Re: Decensoring Hentai with Deep Neural Networks

#64

There should be NSFW examples in the Readme. I saw some practical (NSFW) examples on 4chan and it doesn't seem to work very well. Most hentai doesn't contain simple lines and blobs covering homogenous skin area. There is detail under those areas that the algorithm does not reproduce at all.

Not when Microsoft is about to acquire GitHub!

Microsoft serves abundant amounts of NSFW content on Bing, so I can't see this being a valid concern. I highly doubt Github is going to be censored.

Re: Decensoring Hentai with Deep Neural Networks

#65

Earlier quoted context omitted.

Not when Microsoft is about to acquire GitHub!

Microsoft serves abundant amounts of NSFW content on Bing, so I can't see this being a valid concern. I highly doubt Github is going to be censored.

They don't have anything like SafeSearch in place on GitHub yet.

Re: Decensoring Hentai with Deep Neural Networks

#66

I predict that the porn industry will be the first industry to suffer mass unemployment due to AI. Phase I: Deep learning combined with realistic physics simulation will eventually replace human actors. At some point amateurs can record their movements, then use variational autoencoders to improve their features (bigger boobs, thinner waist, smooth skin) and replace professionals. This is 'Uber phase' for porn and dr…

Porn has always been the earliest adopters of technology. While porn actors may be individually replaced by deep agents, actual porn sales will do fine, especially since extremely targeted pornography will be in everyone's budget. A market for verifiable human pornography will exist alongside the new porn. Also, have we already dubbed virtual ML pornography "deep throat"?

> Also, have we already dubbed virtual ML pornography "deep throat"?

No, that term is already established in the industry and means something else entirely.

Re: Decensoring Hentai with Deep Neural Networks

#67
post #15
post #8

Earlier quoted context omitted.

It's not clear to me that you'd actually need to manually label individual images for training. The work of detecting and marking the censored regions is left to the user, so the model just needs to be good at inpainting. There are probably enough specialized sites to get a decently-sized training set with little effort beyond writing a crawler.

I suggested that to deeppomf a while ago (I was thinking of simply using unlabeled anime images from https://gwern.net/Danbooru2017 with random area deletions to simplify the model & training process as much as possible) and his belief is that because genitals are such a small fraction of any images, and the rest of images vary so much while genitals are a fairly small narrow domain, a generic inpainting/denoising CN…

A neural net that replaces jarring censorship with suspiciously conveniently placed objects? Hilarious...

Re: Decensoring Hentai with Deep Neural Networks

#68
post #18

Earlier quoted context omitted.

It will vary by country. In a lot of Europe and the US the content will still be illegal. In Japan it won’t be. At some point the issue is going to come down to what you want to spend manpower and money fighting, people who actually endanger and harm children, or people who use computers to emulate it. It might not be a very clear cut thing though, because I frankly wouldn’t be shocked to find out that the overlap be…

But how would you outlaw it? A neural network is simply a set of coefficients, would certain coefficient combinations be illegal? It's an interesting question. PS: I am obviously against child pornography, I'm just curious about the implications of being able to generate illegal content and how governments would deal with it.

I think this question is simply a variation of points that have been made clear already in the past - no one legislates against "a set of 1s and 0s", "a set of letters", "a set of ink dots" nor, in this case, "a set of coefficients". Instead, legislation is created/updated for specific end results, and you're done.

For law enforcement purposes, I think it would be similar to a printing press: owning one is not a problem, but inserting printing rolls that look like dollar bills most definitely is.

Re: Decensoring Hentai with Deep Neural Networks

#69

I predict that the porn industry will be the first industry to suffer mass unemployment due to AI. Phase I: Deep learning combined with realistic physics simulation will eventually replace human actors. At some point amateurs can record their movements, then use variational autoencoders to improve their features (bigger boobs, thinner waist, smooth skin) and replace professionals. This is 'Uber phase' for porn and dr…

Porn has always been the earliest adopters of technology. While porn actors may be individually replaced by deep agents, actual porn sales will do fine, especially since extremely targeted pornography will be in everyone's budget. A market for verifiable human pornography will exist alongside the new porn. Also, have we already dubbed virtual ML pornography "deep throat"?

> especially since extremely targeted pornography will be in everyone's budget.

This is why the ML pornography excites me. All my twisted fetishes that nobody caters to can finally be fulfilled.

Re: Decensoring Hentai with Deep Neural Networks

#70

Earlier quoted context omitted.

On the contrary, the examples I saw on /h/ looked pretty impressive for the most part.

Well, I guess we have different definitions of impressive. The de-censored areas did not reproduce genitalia properly at all.

The original illustrators of plenty of hentai do not reproduce genitalia properly at all, if we're being honest.
Post reply on HN