Main work seems to be two years old from the github time stamps?
The generation of images using GANs has improved a lot since then. It would be interesting to see what the latest and greatest methods can do given this data set.
The purpose of this work isn't to generate realistic NSFW images. The purpose is to visually represent what a neural network "looks for" when classifying NSFW content. It is very interesting that these generate images strike the viewer as grotesque even though there isn't anything specific in the images that could be called vulgar. This indicates there is some similarity between how a neural network and our brains classify things.
I wonder where the line is drawn on obscenity for most people? On the NSFW, you can make out some the textures or shapes of different body parts, but none of it is clearly obscene. But there's a general feeling that it is obscene. I wonder where we would draw the line.
> for most people
I think it's really dependent on the individual. Also those 5 judges preside over a nation of only 250M people (1/23rd of the world population)
Depends on who you are asking, which is the problem. Everyone’s frame of reference is different on the matter and I’m convinced the only reason pornography can exist in the US at all is because it would mean changing some part of the First Amendment. Even if I were opposed to pornography, I would be unwilling to open that can of worms.
Not sure why you got downvoted, but Giger would love it I imagine. His were much more explicitly sexual but these are more ambiguously so. And those generated beach scenes were very reminiscent of Dali and Whiteley.
Interesting you posted this a few days ago, because I think it's relevant with yesterdays news of tumblr banning adult content and as a result, lots of images of trivial content (beaches etc) got flagged as porn. See also: https://www.jakeelwes.com/project-MLPorn.html
my recent comments about how important having some systems like this open with self-hosting options for webmasters and user's privacy issues: https://news.ycombinator.com/item?id=18525758
Glad to see this was released and hope some kind of cloud efforts can be used to make it better.
The generation of images using GANs has improved a lot since then. It would be interesting to see what the latest and greatest methods can do given this data set.