Image Synthesis from Yahoo's open_nsfw | Hacker News Reader

Image Synthesis from Yahoo's open_nsfw

open_nsfw.gitlab.io

846 pointsbrakmic168 comments

Open articleView on HN

This is absolutely fascinating.

It's mesmerizing to see this NSFW detection applied in reverse, and it's even more interesting to observe your mind react to the generated images. You can see the sort-of-mons pubis patterns, the maybe-pubic hair, the perhaps-breasts and the suspiciously phallic appendages, complete with realistic colors.

Interestingly, all exposed skin suggests that the training dataset for the NSFW detection was skewed towards caucasians, given how the synthesized images are near-completely devoid of skin tones other than light pink. Perhaps this is a good visual indication of unintentional 'bias' in datasets?

Who wants to volunteer to acquire and organize a several thousand jpg set of "ebony" category porn and submit it? If training material is required, there's undoubtedly a lot of categorized porn for all possible races in the world already, it's just a matter of getting volunteers to do it.

I'm actually really curious whether nudes of dark-skinned people pass the default model at all. Luckily, detector can be seeded to work with a different model, so we can have several of them; but this would be a useful pull-request.

It may be more beneficial to train for just body feature recognition instead of skin tone because last I saw, from the neck down women have only so much variance across body shapes.

We already have a huge amount of homemade porn by people all across the world, but the dataset is not very well labeled compared to professional mainstream porn and that will impact the research.

In other posts I've seen, generative networks produced biases unrelated to what people consider NSFW such as brighter lip color, so I think restricting features related to anything above the neck or below the knees from impacting the CNNs would be helpful.

I would be curious to know what these images would look like if it were trained on a corpus including both the dataset used AND a dataset of hand-drawn pornography, including anime and 'cartoony' porn. I'd also be curious to know how often this sort of network would bar educational content. There is an important difference between nudity and pornography. It's a distinction that I would think would be very difficult to train any automated system on, but I am prepared to be surprised.

glad you like it :) (author here)

when running the experiments, i noticed the neural net responds much more strongly to contrast than color. this is consistent with previous work where the the first layers of a conv net isolate colors in a few filters. i wouldn't read too much into this

This is amazing! I'm particularly pleased to see how it comes out because I remember wanting to do precisely this (generative model plus porn) back in 2000. I had recently seen Geoff Hinton give a talk, and back then I had half my focus on maths and computers and half on art and design. Of course I never got close to doing anything about it :) Maybe it wouldn't have been possible.

I'm not terribly surprised - humans do too. Contrast is well established as being far more important for perceived image 'quality' than color fidelity.

It could also reflect what is most distinguishable. Which is easier for a NN to confidently distinguish: black pubic hair on black skin, or black pubic hair on white skin? Darker nipples on black skin, or darker nipples on white skin? etc You're doing gradient ascent on confidence of classification, not simply trying to find a plausible input, but the maximal input. There's no reason to expect this to be racially unbiased as there are simple objective reasons that higher contrast would be useful. (Similarly, I would not be surprised if a face recognition NN worked better on Europeans rather than Chinese, for the prima facie reason that they have more variable facial features and other aspects like multiple hair colors other than black.)

And since Yahoo needs to detect porn of all races and there's plenty of black porn out there, it would be odd if their porn detector had such a huge gaping hole in it.

> it would be odd if their porn detector had such a huge gaping hole in it

phrasing!

I couldn't resist.

That was deliberate. I couldn't resist either.

Have some mindless up votes Sterling.

> Similarly, I would not be surprised if a face recognition NN worked better on Europeans rather than Chinese, for the prima facie reason that they have more variable facial features and other aspects like multiple hair colors other than black

Hair color, yes, but facial feature variability? I think your perception of more variability in European faces than Chinese ones says more about where/how you grew up than about the actual variability. To someone who grew up in an Asian monoculture, without exposure to media with European faces, all Europeans would look alike too.

Err... I'm not so sure about this. I understand that as you grow up in one community, your brain networks/neurons specialize for distinguishing differences in facial features in that community more than in others. While I haven't seen enough chinese to argue, they have lesser facial feature variations, I can imagine how the algorithm defines facial features would be biased more towards the caucasian/European faces. So from the POV of the "facial feature detection" algorithm, European faces will have more variation in facial features than Chinese faces.

Context: I grew up in southern india and spent most of my early years there, but travelled out to the northern India in the mid twenties. I can now say, I can see facial feature variations in the NE India folks(these folks have facial similarity with chinese) I see around now (I currently live in the south)

>To someone who grew up in an Asian monoculture, without exposure to media with European faces, all Europeans would look alike too.

I'm instead sure about this because of a funny incident that happened to my mother just a few years ago. She was standing near a pizzeria when a chinese waitress stepped out and said (translating from italian, sorry if I make some mistakes):

waitress: "Take away?" my mother: "What?" waitress: "Take away?" my mother:"No, I haven't ordered anything" waitress: "You Italians! You look all alike!"

EDIT: formatting

I imagine it's both.

It's funny how racial bias always comes out in discussion about NN. It all depends on the training data and that's all there is to it. Yet somehow, some people tend to blame the models or the programmers for not correcting this "bias" that is not "supposed" to be there.

If the output of your fancy new NN classifies black people as gorillas it's absolutely the programmers' fault. It is useless for a large swathe of humanity, it creates a PR nightmare, etc.

So nah, it's not funny.

It depends on how you define "programmers' fault". In this case it would more a case of needing more training data to correct the mistake, which may or may not be the programmer's fault.

Of course it can be the programmer's fault. Programming NNs is not just about picking an unbiased sample set of data, you need to help the NN know what it's looking for.

As I've explained in https://news.ycombinator.com/item?id=12759089, different races have different sets of facial features which are used for facial recognition (because they vary more). Now, when programming the NN, what if you only told it to look for the set of facial features you thought was important? It's very easy to let unconscious biases seep into your code.

> different races have different sets of facial features which are used for facial recognition

I thought that «race» is a social construct not a scientific one and thus can't be relied on as a deciding factor in a scientific experiment or endeavor like face detection.

I'm using the term broadly. Google Scholar search "cross race facial recognition" if you want scientific references on this. Or see https://en.wikipedia.org/wiki/Cross-race_effect#References

That's the first sentence of the Wiki article:

"The cross-race effect (sometimes called cross-race bias, other-race bias or own-race bias) refers to the tendency to more easily recognize members of one's own race."

This premise is not well grounded in facts as evidence can show that some people esp. racially unaware, uneducated or desensitized don't fall into this category. To support their claim, they cite a US-only study which can't be sufficient to extrapolate for all humanity as a universal fact since we all know that US society is divided along racial line and it's deeply ingrained into people's minds from childhood.

It's like their caste system and with a highly cultural not scientific element to it.

Color me skeptic about this theory.

Humans perceive and classify race. No-one's disagreeing on that. "Race is a social construct" means that it isn't based 100% on biology or genetics.

Traffic laws are also social constructs. Yet we can build machines to follow them.

Yeah, Humans perceive and classify race (accurately or inaccurately) and that's why it can't be a reliable indicator or cornerstone for any reliable scientific system esp. when taking into account the "fluid" or "conflicting" nature of some of the prevailing definitions of races out there.

> Humans perceive and classify race

No. Some cultures on Earth define "race" with various, conflicting standards. Some don't even have that concept.

Actually, no, _some_ humans in certain places / cultures currently perceive and classify race.

The Roman empire spanned swathes of North Africa and had major centres there and traded with subsaharan Africa. And traded with the 'orient' all the way to China.

Their writings don't make racial classifications. They don't even remark much on skin colour or other features.

Visual features were known, but not considered that significant. And later in the empire people from all over the world became Romans via citizenship and what we now call "race" never factored into it.

"It all depends on the training data"

Exactly.

> I would not be surprised if a face recognition NN worked better on Europeans rather than Chinese, for the prima facie reason that they have more variable facial features and other aspects like multiple hair colors other than black

To people who are used to them. i.e. to you.

I'm Irish, and can tell many Irish accents apart. But I've had people from England not being able to hear the difference between Irish accents, thinking they all just sounded Irish.

I've literally been in a group of Europeans in Africa, and Africans mixing up several of the women there, because they were all relatively tall, slim build, with longish straight brown hair.

That's a classical example of «Outgroup Homogeneity» [1] and I'm surprised that some people are still making these arguments to support their claims.

[1]: https://en.wikipedia.org/wiki/Out-group_homogeneity

It's not a classic example until you can establish that it's an example.

This only means it's hard to judge manually.

> I would not be surprised if a face recognition NN worked better on Europeans rather than Chinese, for the prima facie reason that they have more variable facial features and other aspects like multiple hair colors other than black

This is false. Each individual has a preferred set of facial features that are used to distinguish faces, and this set is usually dependent on our race and/or where we grow up. People of some races look alike to us because they have lesser variability in this set, but may have more variability in a different set of facial features that our mind is not used to looking for.

A face recognition NN would be biased this way only if the training set was biased, or if it was only taught one set of features.

There's a good explanation of this at http://lawcomic.net/guide/?p=3285

> (Similarly, I would not be surprised if a face recognition NN worked better on Europeans rather than Chinese, for the prima facie reason that they have more variable facial features and other aspects like multiple hair colors other than black.)

You might be surprised to learn that to Chinese people, westerners all look more or less the same. To Africans, white people look more or less the same. Our biological hardware is great at distinguishing minute differences in things (especially people) we are very familiar with, while we automatically start generalizing about things we are less familiar with.

For comparison: a metal fan may be able to distinguish a dozen distinctly different styles of Norwegian Black Metal, while to outsiders it's all loud white noise made by people dressed like undead clowns.

> You might be surprised to learn that to Chinese people, westerners all look more or less the same.

Yes, I would definitely be surprised if Chinese people thought a pale-skinned Irish woman with shocking red hair looked more or less the same as an olive skinned mediterranean brunette.

The fact of the matter is, the greatly reduced range of hair and skin color in the Chinese is a legitimate reason to think they look more "the same" than a European/Westerner. There is undisputably greater diversity in appearance in Europe than China.

I encourage you to take a look at a map of skin colour diversity and decide whether Chine or Europe has more variety: https://en.wikipedia.org/wiki/Human_skin_color#/media/File:U...

North to south I think it's pretty clear that skin colour goes from light to much darker than Europe.

The borders of China aren't exactly clear in the image you posted, but here's what I'm seeing:

Europe ranges from 1-14

"Western" ranges from 1-23

China ranges from 12-17

Am I interpreting the image incorrectly? Doesn't it confirm my assertion that China has a smaller range of skin tone?

Edit: The definition of "Western" appears to be very vague, but for reference I referenced at the rough vague shape of the United States to determine the 1-23 figure.

The way I read it almost all of Europe is in the 1-12 range with the exception of Spain. China is 12 to 17 or 12 to 20 depending on how you read the borders. I would eliminate North America from consideration unless we are talking about aboriginals, since it's a melting pot. (That is to say, it goes without saying that it has more variety.)

But my point is only to say that your statement of "greatly reduced range of .. skin color" is a rather exaggerated statement. (For hair color I suppose you might be right ;)

Hard to say due to the lack of precision in the map in the lighter range, but if you think about how incredibly large China is, it stands to reason that your statement might be missing the mark a bit. Especially if you consider other Asian countries. (Which maybe you should, if comparing with "Europe" or "North America"..)

You don't get it. The importance you place on hair and skin color in telling people apart isn't following a set of objective, mathematical criteria, it's a set of priorities based on what you're used to. It is entirely possible someone from a different background thinks of hair- and skin color as irrelevant details, much like you would with a brown and white cow vs a black and white one.

Someone else posted this: http://lawcomic.net/guide/?p=3285

Given a 1 in 10 ratio of blacks in the USA, I expect a similar ratio in porn. It might be different in South America. The algorithmic preference for maximal differences from another comment sounds like a good argument, anyhow.

Re your last point:

Strong pink is probably the most common colour across all people for the most intimate parts of the genitals as AFAIK the inner folds and spread appearance of all types of vulva are pink. All tongues and palms are pink too [AFAIK], they probably feature highly in porn compared to in other images. If you were then looking at pictures of black women, say, and wanted to find the porn images then looking for certain pinks in those images would be a good first pass I feel [you'd probably get lots of false-positives for mouth pictures]. Even those with brown and black skin can flush pink when aroused too, that's got to feed in to the results. [Do men have pigmented glans penis? Not something I've ever really needed to know. [2]]

FWIW https://open_nsfw.gitlab.io/nsfw/prob_0001_1000_10.0_0.04__1... doesn't look especially "caucasian" (ie white European) to me. There seems to be a good mix of pretty coffee-y colours to me that I'd associate with Asia.

>Perhaps this is a good visual indication of unintentional 'bias' in datasets? //

So, maybe it's a colour-of-genitals bias rather than a dataset bias?

[1] http://www.iflscience.com/health-and-medicine/why-are-your-n...

[2] http://www.interestingarticles.com/mens-health-issues/penis-...

> All tongues and palms are pink too

That's not true for palms.

I also think there was a clearer bias towards mushroom shaped objects than O'Keef work. It made me wonder if nude men are considered more pornographic than men.

> It made me wonder if nude men are considered more pornographic than men.

In some places, definitely yes. The laws and regulations imposed by different countries, states, media companies, and other bodies, and how they are sometimes "over interpreted" when being careful not to trip over the rules, do vary but there is some bias towards not showing the penis at least.

In some contexts genitalia is banned but breasts and buttocks are fine leaving more room for a sexual-ish image of woman to be acceptably published. In other contexts genitalia can be show as long as it isn't in a obviously aroused state, but other shows of arousal (blushing, other body language, suggestive word choice, ...) are fine, which will likely bias against male genitalia being publishable unless close up (and in contexts where the question is even raised a close-up shot is less likely).

(Massive generalism here, but...) there is also a general opinion that the use of the female form can be defended as artistic more than the male - though I couldn't tell you if this is a symptom of other cultural biases or a cause of some of them.

Attitudes to sexual preference are going to factor in too: those anti-homosexual opinions/feelings (conscious or otherwise) are more likely to react strongly to images of male sexuality than female (the old joke "I support homosexuality as long as both the chicks are hot" has some basis in reality). This is more common in men in my experience, especially man of certain age ranges and backgrounds which are currently over represented in parts of the "traditional" media industry and other seats of influence, making the bias more visible.

All that and a number of other interrelated factors I don't have time to type out make it less common to see a full-body or otherwise genitalia-containing shot of a naked male than one of a naked female - so when one is shown it is more likely to illicit a stronger reaction.

(I avoided using examples of differences in the consumption of pornographic output between different sexualities, as there is a lot of reporting bias in studies in that area both in terms of what questions the studies ask, how truthful the subjects answer, and what gets selected for publication)

Some of these images and those from similar projects could be in an art gallery. They are art; provoking original, emotional, responses.

Most people hear about self-driving cars, but not about the fact that machines have already begun to emulate human creativity in the most intimate way. For a while, this secret assault on our uniqueness will stay among us.

He built an H.R. Giger [0] generator, essentially.

[0] https://en.wikipedia.org/wiki/Penis_Landscape

I thought the images looked mildly interesting but lack a certain sense of Giger.

With a bunch of Giger images plus a bunch of porn as input the output would probably be fairly indistinguishable from Giger originals.

My immediate thought was that they looked like the work of Salvador Dalí

Sounds like the premise for the strangest Viagra ad ever.

Calm your horses. Generative nonsense have been already explored in literature in 90s. So for me it does not produce any new feeling.

Now for a newer generation playing with a real ding is perhaps a little more intriguing.

Actually, no. Nothing produced in the 90's is the same as what's been done in the last few years. But please cite your best work and we'll have a look.

Regarding your dick joke, it's not really relevant. This is not interesting because it is pornographic.

The work of 90s as I know it (sorry, no references, I was quite young then, do not care to loop up at the moment), was taking some corpus (and not an English one) and letting a generative algorithm run on it. This was not a research but art, literature, and as such, actually could have been very regional.

Of course the current state of art is on the different level but in basic principles not very different and as such not very surprising as an form of art.

PS. it does not imply that no interesting results may grow out of this.

I disagree, because the machine is still using human generated input for creating these images. Not to take away from what's been created here because it's interesting to look at, but an image filter could be considered an attack on our uniqueness because it takes an existing human art work and can potentially create something more interesting, but most people wouldn't consider it an assault.

I get this is a "neural net", which makes people think these paintings are created by the terminator, but these images still have very real human inputs which is why we can relate to them.

What human input? I mean the pictures were taken by humans. But they weren't drawn by humans, they weren't meant to be art, and no one intended the output of the network to look the way it does. There is very little human input here.

If I pick a shell I find beautiful from the shore, did the beach create the shell, or did I?

Computer algorithms produce a lot of trash. Just because the human didn't wield a brush, as long as we're seeing filtered and tuned results, a human had a hand in creating these.

Yes humans were instrumental, but we can acknowledge that while still accepting their role as separate from artist: http://www.nytimes.com/1996/06/30/movies/how-creative-artist...

> already begun to emulate human creativity in the most intimate way

I'd argue that the creative entity here is @gabeeegoooh. The machine didn't make the (provocative) decision to synthesize NSFW images - that artistic stroke was painted by the fingers and mind that coded up the adversarial neural net.

One way to look at @gabeeegoooh's role is as muse; He provided the "inspiration" for the creativity.

Definitely. If creativity were merely the creation of interesting images, we'd have to call nature far more creative than any human artist. The Grand Canyon is beautiful, but we don't put up a plaque giving the Colorado River credit for its thoughtful use of form and color.

Calling the neural network creative is like crediting Van Gogh's brushes for Van Gogh's brushwork. It's a possible use for the word, but it's very different use than the one normally meant.

Van Gogh's brushes were simple tools used to render a direct expression of his intent.

These networks are more akin to a child. You teach her to paint, give her input and feedback. The result may be something emotionally moving that you could have not even conceived of.

Your opinion and mine are both surely subjective. However, it seems it will only become more difficult to split the hairs required to argue that machine learning creativity is fundamentally different from our own.

Brushes are not pure transmitters of intent the way that, say, Photoshop is. Brushes (and paints and most artistic tools) have an artistic character that makes its own contribution to the artwork. I also think you ignore the interplay between medium and artist, the way intent evolves both during the course of an artist's careeer and during the production of a single work.

I don't think I'm hair-splitting here. Creativity is something conscious entities do. Once we have conscious machines, they could well be creative. But until then, the creativity is the artists, even if they are using generative systems to create art.

And that's nothing new. Every landscape painter and photographer is using a generative system, nature, to find interesting things worth turning into art. Giving the neural networks credit is akin to crediting the haystacks rather than Monet.

More explicitly generative art, where humans set up systems to produce works, goes back hundreds of years. That interesting patterns can emerge out of randomness plus systems is obvious to anybody who has played Yahtzee. That we use more dice and more complicated systems does not change where authorship lies. Bigger systems may produce more novel outputs, but novelty and creativity aren't the same thing.

So, in what way would I be wrong if I called you a generative system?

I don't see any connection to consciousness whatsoever.

Imagine in the year 1900 a black box existed with a slot to input paper on one side and output paper on the other. If all physics papers available at the time were fed into it, and out popped the special theory of relativity, how is that not creativity?

Einstein is a black box that reads physics papers and poops theorems? Sign me up!

You might not see any connection, but it is nonetheless how people use the word. Human language is mainly about humans and their relation to the world.

If your black box contains Einstein, then yes, it's a creative act. If it contains an artificial consciousness, then yes, it's a creative act. But if it contains infinite monkeys typing randomly into LaTeX producing infinite physics papers, then no, that's not creativity, just the generation of novelty.

I could call a red rock-face "angry", and I could call a mountain lake "tranquil". But that's projecting emotional states onto the external world. Calling dice creative is projecting a cognitive state onto the world. Sure, it works fine as metaphor. But it's just analogy, not homology.

I've enjoyed your comments, they've helped me clarify my thoughts.

It seems we disagree on what creativity is. I strongly believe humans/consciousness are not essential elements in it. For starters:

1). No dictionary seems to link creativity and humanity

2). Einstein's insights are considered some of the greatest examples of original thinking ever, and would not be diminished, or less beneficial to society, if conjured by a black box.

3). By your definition, a machine could out do a human by a thousand fold in every possible task and still not be creative. I don't think there's any precedent or foundation that would make that a tenable position.

Glad to hear it. And I agree that we disagree on what creativity is. As to your points:

Dictionaries are written by humans. The human is always implied. One way to tell is by looking at synonyms: inventiveness, imagination, innovation, innovativeness, originality, individuality. All very huuman words.

If the theory of relativity came from a black box, then sure, it would still be a good theory, and we'd still benefit from it. That does not make the black box creative.

Conscious machines could indeed be creative. We are, after all, conscious machines. But creativity involves a shift in understanding, and nonconscious machines don't understand.

And the curation of the images, the selection of which to feature, is also a very important aspect. He didn't include some images the system did or could generate. While he personally may not have, it is absolutely possible that such a decision was made in order to communicate some idea. If you take a random object, and place it in a museum, that can be art simply by virtue of the fact that you are attempting to communicate something. Art is, in my opinion at least, an attempt to communicate (typically supra-lingual ideas, concepts, feelings, etc).

I am always blown away by how eerily similar these generated NN images are to the visuals experienced under psychedelic drugs. Moreso than any artist's depiction (and there have been plenty of those)... they just have the same "feel". Which of course leads one to the inescapable idea that there is a fundamental relationship here.

Maybe dreams and hallucinations involve your brain running recognition networks in reverse to generate stimuli?

I've hypothesised something like this. Or rather dreams could be generative based on random input (or random 'memory key frames') rather than real input. This assumes that mostly what we 'see' when awake is actually generated by the brain based on incomplete and noisy sampling of the real world.

This is basically the exact hypothesis of a TED talk I saw by cognitive scientist Donald Hoffman.

https://www.youtube.com/watch?v=oYp5XuGYqqY

fascinating stuff

I think that's pretty much the standard theory nowadays.

If I understand my pop-science correctly, our subjective visual field is actually made of images synthesised from objects we have recognised.

When we are awake, this mostly matches the real world because we select those objects to match data coming from the eyes. But when we dream or hallucinate, the synthesis circuits fire based on other inputs -- such as noise.

Indeed, and let's not forget the extended example of the video of random noise fed into Google's Deep Dream system: https://youtu.be/SCE-QeDfXtA

This was absolutely fascinating. Thanks for sharing.

Some of the more abstract images at the beginning really remind me of Beksiński's paintings. (some NSFW, but good, dark art overall) https://art.vniz.net/en/beksinski/ There's just enough of abstract ideas and randomly included genitalia.

(now I really wish someone did a Beksiński + photos mixer... there's ~240 samples just on that site)

Remind me of the paintings of H. R. Giger, the designer of the creature in the Alien franchise.

Google "giger nsfw" if you want to see his more sexual creations.

Giger was the first thing that came to mind for me, too.

The NSFW images actually reminded me of Keith Haring's work, given the not-so-subtle phalluses in all the images.

Holy shit! It's like Salvador Dali, Gustav Klimt, HR Giger, and Frank Frazetta had a love child and locked it in a closet. Thanks for the link.

A movie called Ostatnia Rodzina (Last Family) about Beksiński family was recently released in Poland (https://www.youtube.com/watch?v=XfFt9RfO9Bc), definitely one of the best movies I've seen this year. Unfortunately I don't know if it will screen outside Poland and UK.

There's a fair bit of Hieronymus Bosch in there too.

Anyone know where I can get prints of these? Some of them are insanely good.

There's lots on http://www.boszart.pl/pl/14-zdzislaw-beksinski- - the prices should be pretty good after exchange.

You'll have to translate from polish, but they do send internationally. The email contract is at art@bosz.com.pl

Fun fact: with prints, without seeing the original pictures in his home town you miss out on his notes of what substances did he ingest while painting. (sometimes attached or on the frames if I remember correctly)

Damn, thanks for the response! I've been to Poland before and I intend to return, so I'll add his home to my "to-visit" list.

I'll have 2 tickets to the dick concert thanks.

Those ones were hilarious.

We're witnessing the beginning of an entirely new form of pornography. I can easily imaging a XYZ Porn website adding an "Artificial" or "Neural Dream" porn section.

Reminds me of http://www.scp-wiki.net/scp-1004

Wasn't this the sort of thing the plebs were distracted by from the porn division in 1984?

They have automated the surrealist movement. Which goes pretty much directly against the philosophy underlying the surrealist movement. Which the actual people involved with it would probably approve of, as they mostly all moved on from it anyway.

This is one of the most disturbing things I couldn't stop reading.

Agreed. These creepy images make me wonder if some neural net-based text generator could optimize against algorithms detecting both readable English and scary/traumatic/intense/whatever prose.. and produce mental imagery darker than any human author has.

It wouldn't maximize fear response in humans, only in the neural net that results from the training set.

If we get to where we can simulate human neural nets (and identify the "scary" neurons) then we could perform gradient descent against those. That's an interesting concept. I don't know if that's been visited in SF, but it has potential. Imagine a set of not-so-random squiggles that causes uncontrollable fear to its beholder.

https://en.wikipedia.org/wiki/BLIT_(short_story) might be of interest to you.

Man, Snow Crash basically sounds like it stole that idea.

There was also an Outer Limits episode with a similar theme if I recall correctly. There is a lot of cross-pollination of ideas, and I think that's a good thing.

Or a lethally funny joke

I believe this is actually a possibility now.

So when I was back in high school I was playing around with one of the Kai art programs (I forget which one actually) that let you generate different patterns. I couldn't explain why but the patterns that were kind of regular-ish and curvy/abstract (anything that looked like cells under a microscope or those with a honeycomb-like structure) made me deeply, anxiously, almost panic-attack level uncomfortable. This startled me when it happened the first time. There was nothing at all I could logically connect it to.

The best way I can describe it is that I get that deep stomach-fluttery feeling that comes with being really nervous and apprehensive, but without any of the emotional content of anxiety at all.

I forgot about that experience for a long time, until the first Google 'Deep Dream' images were released. Turns out that the Internet has given this a name - trypophobia (fear of holes). The regular, not-quite-identifiable quality to the images is what seems to do it to me.

Even now, my stomach is fluttering like I'm about to go on a scary roller coaster. Again though, this comes with no emotional panic. I'm not upset by it, it isn't self-reinforcing like anxiety is. It's just a really odd feeling. A very strong 'get away from this' feeling.

I've read the hypothesis that this comes from an instinct to avoid danger, such as beehives and poisonous creatures and that makes a lot of sense (not totally sure from memory though so I might be getting it wrong; I need to find the study I'm thinking of).

So I don't doubt that at least for some people it's possible to synthesize images that provoke a strong physiological response that could make them flee.

Gut instinct wants to say that it's a bit of an uncanny valley experience as well. Our brains are wired to recognize patterns, and seeing something very organic, but also very mechanical just sets off those deep instinctual alarms that something's not quite right.

I also have trypophobia.

Sort of related: Instead of visually producing fear, you can also do it with sound.

Research "Binaural beats: Gates of Hades" I literally threw my headphones off at like 25secs.

I know you could literally mix trypophobia images and Gates of Hades to produce a waterboarding like response.

Is this the track by I-Doser on Spotify? I listened to it with decent headphones and it just seemed like ambient noise with some chanting in the background. Should I have listened to it in a certain setting, or with eyes closed, or while looking at something?

How many outputs need to get evaluated for scariness? Mechanical Turk might be a feasible black-box replacement.

For NSFW stories / erotica, there is Eros Errors (http://www.eroserrors.com/). Generated using Karpathy's char-rnn.

By popular demand, I've added more pictures!

https://open_nsfw.gitlab.io/more.html

At first i thought WTF am I looking at? But actually they are quite beautiful in an abstract sort of way.

can the process generate hi-res images?

Seconded! I'd love to have some in really high res (I dunno what for, but I'd love to take a closer look). However, even medium res would be really cool, IMO the generated pictures just a tad too low resolution to appreciate carefully.

Is it a compute/processing power thing? Does it take a lot more time to generate an image at say 5x resolution?

Also what I wondered about, quite a few of the images seem to have a sort of "canvas" like texture on them. Any idea why that is? Would a lot of the source images have canvas textures? Is it an artifact of the algorithm? It kinda feels like a highpass (edge enhance) filter as well. Or maybe they're just JPG artifacts?

the artifacts are explained in detail here

http://distill.pub/2016/deconv-checkerboard/

more compute will enable higher resolutions and less artifacts, but it means training a bigger GAN

Could this be used to insert a "subliminal" touch to an image?

Eg you make an ad that looks innocent, but that would fool your brain into thinking it's sexual if you just scan the page containing it?

Is there any evidence such marketing works?

Advertisement is not only marketing, it's also art. Even if ad agencies will fail to convince some porn site or condom producer to use such ad to generate news, they still can create social ad and submit it to advertising festivals.

Not to mention the potential to generate free press coverage from "controversy."

Has anyone ever thought about using all of reddits porn subs for machine learning? There must be 10s (or even 100s) of thousand of images (kind of) neatly organized by gender, boob size, ass size, skin color, age, ...

Or any porn website.

The advantage of reddit is that you can also get tons of cats, dogs, cars, landscapes, food, ... all from the same place with the same scraper.

If you want fine-grained tagging/categorization, you would be even better off with the -boorus.

Quick plug for my map of NSFW Reddit: http://electronsoup.net/nsfw_subreddits/

Unusable on mobile. In portrait, the sidebar covers everything; in landscape, it still covers half the viewport, and touch panning works but resets the position of the graph to what I assume is center or origin on every drag.

Hate to seem harsh, but I felt it worth mentioning, since it's why I'm unable to give any more substantive feedback on the project.

Not at all designed to be compatible with mobile.

Cool visualisation, but it doesn't have links to the subreddits?

I thought of doing this but my bottleneck is scraping the images since I don't have terrabtytes of bandwidth and imgur would likely rate limit me. But given the data set my plan is/was to extract tags by mapping images to the sub Reddit they're posted in then categorize based on that

Contabo are 100Mbit but unmetered, ridiculously cheap.

Reddit's main issue is that you can only do one call (via OAuth) per second. I recommend using the new(ish) "thing" call where you can get up to 100 things at once, or alternatively rockets.cc, which occasionally falls apart but is pretty good

There's no reason to scrape Reddit for the image URLs when you can download all NSFW image links on Reddit via BigQuery.

Beware, many subs mark links as NSFW to indicate spoilers, rather than pornographic content.

Obviously you can filter on specific subreddits. :P

They've figured out how to synthesize OMNI magazine covers.

Shoulda called it "Deep Dicks" or something.

I would LOVE to see what happens if you feed it clearly NSFW images as the source, and let the network optimize for SFW instead

It may look surprisingly similar: https://codewords.recurse.com/issues/five/why-do-neural-netw...

Fascinating, I'm curious to see if any of the tech press or even mainstream press pick up on this. If they do, will they pixelate the sample images??

I'm seeing Roger Dean (of the Yes album covers) or maybe "Heavy Metal".

Whatever Piaget stage prog rock is, AI has reached it.

I wonder if it would make it hard to find this project via Yahoo's search engine. That would be sweetly ironic.

If the author is reading this, per se means by itself, on its own. I'm not reading per say in its stead for the first time today.

It shouldn't be surprising that many misheard words survived in a time when there was no widespread frequent exchange of written language and no writing standards or before that, when hardly anyone could even read. I feel this severely complicated our languages.

This is slightly on topic as well, because Natural Language processing has to deal with that now.

going one step further with the nitpicking, just because I am at it, the per-say (or indeed, per se) is only a filler in that sentence, like really or very often are, really though.

"Not surprisingly, the results of the optimization are clearly pornographic."

Just about spit out my food.

So what about optimizing NSFW originals to make them appear "SFW"? What would such a thing look like? Presumably the skin tones would be the first to go.

https://github.com/tjwei/play_nsfw I have some experiments along this line.

Georgia O'Keefe has already contacted her attorneys...

... or rather, her estate has.

I was reminded more of Salvador Dali's work, particularly The Great Masturbator.

H.R Gigerish.

It seems possible to put other, not-porn images of people into the hopper and spit out an endless stream of perturbing, semi-pornographic trolling. That will probably happen, despite it being an awful idea, and it could even become commodified.

The repository is here: https://gitlab.com/open_nsfw/open_nsfw.gitlab.io/tree/master

Is there are a good validation set of subtle examples of something being NSFW and SFW?

Especially considering that the subject is mostly defined as I do not know to how to define it but I definitely know it when I see it.

OK - does anyone know, how close are we to these trained neural nets passing the turing test for human creativity? Because I feel like we're going to pass it in my lifetime.

Hang prints in a gallery and let's find out.

Sometimes the amount of time and effort we waste on enforcing outdated morals on other people astonishes me.

Still, we get abstract genitals as a side effect.

The output looks like something Harry would have run across going through the apartments in Silent Hill 2.

i don't see any pictures in the link.

author here, i just realized this site doesn't work on mobile. I used javascript to generate the tables of images. While I blunder through trying to fix this, you can view the pictures here

https://gitlab.com/open_nsfw/open_nsfw.gitlab.io/tree/master...

Fixed!

If you were on safari before, it should be working now.

Just confirmed the fixes are live.

Why exactly are you using JavaScript here and not a simple HTML table or list?

a fair point.

I imagine he was trying to save time by writing code that makes the tables, rather than manually making them.

Perhaps predictably, Facebook is censoring the URL by refusing to let you post it.

Seems to have bugs. I dont understand whats NSFW about Ted Nugent at a rock concert

similar results using a smaller set of inputs, 20 years ago: http://draves.org/fuse/

so this could be used for an automatic nudity free chatroulette

Talk about "...You'll know it when you see it..."

Very HR Giger esque. Somehow more horrifying than the original.

So that's what they are doing at Yahoo these days.

It's... it's a robot Dali Giger.

I love living in the future.

ah ... so it doesn't dream of doggies then .....

Reply on news.ycombinator.com