We could create a Pedo detector, Holla Forums

That would be 6 million times easier to do too. You could probably make a decent Jew detector relying on the curvature of the nose alone, given a good side view. Although there are many other phenotypical indicators to look for- epicanthic fold, negrid-ish hair, frog mouth, bat ears, balding, evil smell, either hysterically high-pitched or creepily low voice, beady black eyes, 'Rabbi' facial hair texture/pattern.

It's probably just whether they have a masculine face with a strong bone structure, or a weak, nu-male face.

No, all women are bisexual. Or omnisexual, I guess. Hook them up to a machine that measures blood flow to erogenous zones and show them footage of anything having sex. Doesn't matter if it's straight, gay dudes, gay women, dogs, monkeys, whatever. They get aroused, no matter what they think or say their sexual preference is.

Men, on the other hand, either get aroused by viewing straight or gay sex. One or the other, not both. This doesn't necessarily have to match with what they think or say their sexual preference is. The study showed that some gay men are biological, some are actually straight but conditioned to be gay. Women, however, have an arousal response to viewing pretty much any sexual activity.

But the thing is, gay isn't genetic like they try to spin with pseudoscience news like these. Gayness is cultural and learned. So you can have a masculine guy in a degenerate setting who slips into a fetish like that and you can have someone with soft features who don't.

You really get a certain special sense from perverts, but it's hard to pinpoint exactly what it is. Maybe it's some sort of spiritual thing.

For the vast majority, this is true, especially now. But it is possible for someone to be born with a predisposition to homosexuality due to environmental factors during prenatal development, or the mother being a piece of shit and ingesting anything she feels like. These men generally tend to have more "Female" brain matter, and have a harder time with spacial reasoning.

Perhaps. I've never seen any real evidence for it outside gay lobby "science." But rare freak of nature mutants aren't really something to build any generalized knowledge on anyways.

Try 16-18. Women’s looks go downhill really fast.

Nah I’m just not a faggot.

...

There aren't much physical traits on this that I can think of other than external factors ie short hair, morbid obesity and others. But now that you mention it there is a bit of a fishmouth among dykes but that's a common SJW trait in general. They keep their mouths open all the time because they're always thinking about pussy.

But other than that, the foolproof trait to look for is the ego. Show me a woman with a massive ego, including but not limited to career women, and I'll show you a lesbian. The massive egos of some of these women is what turns them away from men, because that's "below them" and "oppressive and patriarchal". Resting bitch face is part of it too, and by the way resting bitch face is a meme from bitches to hamster that they aren't actually a bitch. Usually if someone is used to making an upset face, then they're actually pissed most of the time, because if they are legitimately not angry then they will make an effort to not look angry. Source: I'm a guy and I "have" it.

True enough. I got all this though from a human sexuality class, which I took on a lark, but ended up being the best elective ever. The professor was a fucking boss. When it got to all the homosexuality stuff, he triggered the fuck out of half the class and said "Not a single man in this auditorium is bisexual. If that bothers you, get out, because this is a scientific fact, and if you choose not to believe in scientific data, you don't belong anywhere near academia". It was great. I looked him up a few years ago. Replaced by some literal nigger faggot, and I'm not at all surprised.

I think some gays are so because of genetic causes, but not all. However, this "gay look" does not imply genetic causes. Face changes due to environment, too, specially wrinkles, and drug-like substances alter the brain (affecting gestures), and the body.

There's a limit to how much can happen and why facial recognition software has become so good.

I have a better idea. Kys.

Well it's not like he could get into any more trouble now. If you could look him up then I assume we could too and find some of his material? What's his name?

There is evidence that child abuse (physical and/or sexual), and be raised in a broken family (without a good male figure) can turn people homosexual when they become adult. Adult faces change much less, though.

About the "vibes", there are studies regarding rapists, in which it was shown that most male subjects were succesfully able to differentiate the faces of rapists from normal men from a list of suspects. However, women could not differentiate them at all. The study concluded that this "advantage" was what allowed them to approach their unsuspecting victims.

There is definitely something there, and not prejudices showing.

natural irl family photos would be much preferable
mugshots are very peculiar, with unnatural tense or absent expression, the learning algorithm may well be thrown off-tracks with that

David Puts. Turns out he didn't get shitcanned, they just shifted around what he teaches and replaced his "human sexuality" class with a nigger. His field is anthropology, and obviously nowadays you can't have an anthropologist teaching about sexuality. Pretty solid dude, at least when I attended. Nothing but respect for the guy's work.

Kek that's probably what the yids think of him
I'll be sure to look into some of his work.

Also
He would be more suitable in biology or veterinary sciences.

There were studies on that too. Men simply have better threat detection than women as a necessity of evolutionary survival. For example, male brains have more grey matter by around IIRC 15% than women and a larger part of the lobe that controls for threat analysis and calculation. It's a shame our ancestors didn't select for women with better threat detection ability, maybe then they wouldn't be so easily fooled by Chad or muh poor refugees.

As for homosexual abuse,

Abused "More Likely to Be Gay"
stuff.co.nz/national/3949087/Abused-more-likely-to-be-gay

Homosexual men are more likely to have been abused by their partners than straight men.
takimag.com/article/the_straight_dope_on_homosexuality_elizabeth_mccaw/print

Gay Youths 148% More Likely to Be Physically Abused in Relationships: DOJ-Funded Report
lifesitenews.com/news/gay-youths-148-more-likely-to-be-physically-abused-in-relationships-doj-fun

Yeah, and women get 'aroused' when they're being raped too. I suppose that makes them rapesexual in your estimation.

fucking jij

tits or gtfo, don't you know this by now?

Except he's right, and you are too in your irony. Personally, I think women's brains are built to function in a harem. Monogamy was a creation of man to build more stable societies.

Shitskin maddox made such a hypothesis years ago.
thebestpageintheuniverse.net/c.cgi?u=spot_the_pedo

PEDOS ARE THE FUTURE!!!!!!!

OP

If you need a database of child sex crimes and mugshots… familywatchdog.us is a good place to start.

It does.

archive.is/w80HZ
And the most common trait I see, one bulging eye larger than the other. In fact, there are two subtle examples in OP's pics, Acosta and Nye.

Source please on rapist identification. That's a fucking gold mine if true.

Jesus

*average

Pedo smirk is real. Maddox wasn't the first to notice. I think pedos develop this kind of 'naughty smirk' because part of their pedo-sexuality is its forbidden aspect.

Like kikebook claiming they had to shit down their ai because it was speaking in a secret language which turned out to be simply repeating one word over and over again.

And then there's "lipstick lesbians" aka women who kiss and touch other pretty women because it gets guys' attention. They get added to homosexual sampling quite often.

It's a smirk that's a combination of the smirk that adults give to children to get the children to smile and a devious smirk knowing they're getting away with something disgusting.

wrong, all males swing both way, they might be only attracted to males and they don't seek out females, but females are closer to them and when drunk they advance ion their gay friend if they can't get anything that night, everytime

also, on topic,
pedos have a forced smile, that is why they look weird. I can only speculate why they force their smiles, but it might be:

Stop fantasizing and start kys. Why wait, start today! Fucking faggot.

Yes. You could create a database with tens of thousands of mugshots of scum convicted of sex crimes against children. Statistically is that enough with so many people? No, but add in some demographic data and other physical traits and you might get somewhere. I say, kill all the ones that are out of prison who have been convicted two or more times of such crimes. You'd be surprised how many thousands that is. These monsters are genetically defective. As such there's no curing them. They have to be put down like a rabid dog or jew.

My understanding of the technology involved is nigger tier so without proper direction I will be unable to help. But should you succeed in doing this there will be a huge outcry when its found that jews are found to be disproportionately represented.

Make this happen anons.

What absolute nonsense.
Pedos are mostly jews and/or people who were abused in childhood.
You will find that pedo-radar is also jewdar.

Fishmouth is from massive alcohol consumption in infancy

Interesting, so its not just OP

"Your time is up, faggot!"
HAL-9000

Plenty of projecting degenerates.

Storytime.
Her face reaction was amazing and i can still remember it because back then i didn't understand why she reacted that way.

I thought he was Armenian and therefore white?

Holy shit, user, that is some weapons-grade autism

I don't understand

I've never read anything along these lines but, I've always thought that female arousal during rape was a defense mechanism so the vagina doesn't get damaged as much as it would be when dry.

Fucking kill yourself faggot.

Hybristophilia, newfag.

Rape babies grow up to love rape. Who knew?

You have what fags called gaydar.

You can apply whatever makes up "gaydar" to pedos as well I believe.

One theory I have is that our sexualities are such an important part of being human that it manifests itself in all areas of life. It's about the little things. The actions no matter how small, our interests and how we present ourselves is all an expression of who we want to fuck.

Normal straight men all act a certain way. We act "normal" but gays don't. They have different priorities and therefore it manifests in different tells. This goes being the limp wrist and lisp. It's the way they stand, walk and interact with others. It's where there eyes go in a room full of people. This is how fags can tell who is a fag and who is not.

There is a reason a pedo is "creepy". It's because he acts off. Or speaks off. Or carries himself in an "off" way that doesn't signal that he is trying to impress the opposite sex the way a normal man does. The drive just isn't there and it manifests itself in many different subtle ways that you can't detect so most people just say "that guy is a creep"

This is all subconscious stuff btw. Men and women both peacock subconsciously to attract the opposite sex even when they aren't actually looking for sex (married, celibate, etc.).

I think the real mistake is assuming that female arousal is anything at all like male arousal. They're not aroused because they're attracted to someone, or because they're horny, or whatever other psychological phenomenon you can think of. They're aroused because they're fucking animals and it just happens without even being something that crosses their mind. Have you ever seen two dogs fuck? They're doing it without even understanding what's going on. They seem to have no control over their own bodies at all.

I think it is more like a personality type. What attracts their focus, what motivates them, their interests, how they behave around certain types of other people (pulling your dick out in front of children is a red flag here).

The homo females are more attractive than the straight females.

No women are really lesbians, they're just attention seeking whores, who do things with other women to attract and alpha man to fuck the both of them at the same time. Ya.

No, you're just a faggot and can't identify latent queers.

Apt-get install python3 python-setup-tools python-dev build-essential python-pip
Pip install scikit-learn tensorflow keras

You might need PIL too. Keras is stupid powerful. Great documentation.

Bump for interest. Keep the thread alive until I get back from work. If any Anons have a good source for training data,
please post (both pedos and non are needed). If it detects Joe Biden, we'll know it's working.

OPERATION EIGENJEWS

Convolutional neural network with 3 or 4 hidden layers (convolution+pooling)
You would just have to sample a database with a lot of pedo faces and train the net

There was a fairly recent study that found that people could guess people's names by their faces alone, well above random chance. They conducted this study in (((Israel))) but on another country, too. They made lists of local names and asked the subjects to guess the name of the face presented to them. Cross-culturally, it was random chance, but locals guessing other locals names were above chance. That's the long and the short of it, can't be assed to search for the source right now.

The researchers reasoning for this was that because people are given names, it becomes a part of their personality, and being called that name subtly changes the features of your face. So basically, nothing short of meme magic. So based on this, I would say that there is something to this logic that OP presented, I could see it tie into the nature + nurture mechanic of life. One is genes and one is environment. Genes are fairly straightforward but the environment shapes how you perceive reality on a much bigger scale. You can't live life in a void after all. It would be interesting to see what this approach would yield, especially concerning if there are "latent" facial traits that can predict sexual orientation, but the results won't ever see the academical light of day. It will only ever stay a curiosity to consider.
If it was verified, it opens up the floodgates of a new scale of shit-flinging. It would mean doling out vigilante justice for those who deserve it and those who are just unlucky to be born to look like a pedo, same reason why race studies are so taboo. Can you imagine being hounded and excluded from everything regarding work and social situations, just because you share facial traits with a pedophile, while not actually being one? Same reason why a broken clock is right twice a day or sometimes you end up seeing doppelgangers of yourself in countries on the other side of the equator, some things are just up to random chance.

Of course, one other reliable way to gauge people is to be around them. Text can only convey very little in terms of gestures and tone, same as with faces. We only get one part of the bigger picture if we concentrate ONLY on dismantling facial traits, since we need to know what other defining traits, if any, are to be found in other areas of life regarding sexuality.
Everyone knows there are subtle differences, but no one wants to verify the claims and add fuel to the fire. It's better to keep people guessing and let them live their lives in something resembling balance, at least I can imagine that being the legal/ethical approach.

Your study sounds like retarded gay kike bullshit.

You should kill yourself for being a gullible faggot.

Eat shit you fucking nigger whore child mongrel cocksuking piece of bloody AIDS ridden faggot shit.

I already have one, usually when it goes of it does something like this ((()))

While I was studying body language, I came across some of the works of Paul (((Ekman))). He did a study of facial analysis to detect emotions and lying, leading to the development of the FACS, Facial Action Coding System which acts like a map of possible human expressions.
I'm not sure if it was him, but he or one of his colleagues did a facial survey of races around the world. One distinct note I remember reading is the ability to discern a culture's predilection for boy sex. It outlined various features to look for, lines and facial morphologies. I have been trying to find it since Pizzagate broke, with no luck. Original sources were put out in the 70's and don't want to show up on web searches.

Dumb larp thread. The way to tell if someone is a faggot or not is to look at two things:
Same for pedophiles

There are four types of lesbians though:

The last two are probably what's skewing the results toward looking younger and more attractive. The truth is simply that "lesbians" as in homosexual exclusive women aren't "real" but the behavior of fucking around or just playing pretend and self labeling is.

REALLY MAKES YOU THINK HUH?

Actually, for a computer learning algorithm, you want all other factors held constant. It doesn't matter if the lighting is awkard as long as that lighting is consistent. Pulling family photos would lead to huge inconsistencies in the image statistics and would require a much more robust classification algorithm. You notice the compiled pictures in the first OP image? All those photos are taken against identical backgrounds, with identical lighting conditions, centered exactly on their faces, etc.

Oh hey, I've heard of this being done before. There's an entire obscure science that got formed out of identifying facial clues to read minds.
en.wikipedia.org/wiki/Facial_Action_Coding_System

I used to believe you, but I sometimes find bulldykes in my herd. It is about a 1 in 500 chance. They have big muscles all over like a bull and will try to dry hump the other cows. They attempt to fight off the real bulls but allways get their asses kicked anyway. I sell them whenever I find one because they don't always get bred, which is not optimal. I suspect that it is some type of hormonal disorder in mammals that makes them that way. Maybe too much androgen.

Not really, you should head back to cuckchan.

ANYONE THAT DOES NOT IMPREGNATE 14 YEAR OLD WHITE FEMALES IS SUPPORTING WHITE GENOCIDE

For one distinguish behavior of a pedophile is that when driven to corner, he temporary flees from one thread to another. Coward

...

have you ever looked at your local sex offender registry? all your "pedo detector" would do is look for mexicans, niggers and white men with mustaches

Don't forget trannies.

reddit spacing

"get em while they're young"
The whole muslim thing of subverting the youth not by bullets, but by pedophilia

What are you even doing here?

Reported.

there arent enough jews or trannies in the population to change the composite image. Even if 110% of jews and trannies were pedophiles, thats still like 1% of total pedos

this is known as physiognomy.
it fell out of fashion in the 20th century but is making a comeback, though its mostly shitlord science now so people go out of their way to discredit it.

Yeah, reported for being mentally defective

If any, the results of this research will lead to more accurate portraits of the prophet.

14 is perfectly marriageable age. It's even AoC in most of Europe.

user here

So nobody has a dataset? I'm seeing a lot of posts about generalizations of human sexuality. Not a lot of productivity. A reasonably accurate classifier could blow apart careers and identify high value targets.

NSOPR.gov has searches by zipcode, and getting data from each state is different. NY has fucking captchas. I can write state specific scrapers, but if anyone has a better source, I would MUCH rather not spend a week or more just getting enough pics for a training set.

I will look for datasets this weekend. We will probably need to categorize them by men and women, and probably races too.

Ive always noticed lesbians have super dialted pupils, all the time, compared to women who only have dilated pupils when they are turned on/ being sexed

lip and nose peircings

What are you a queer?

but of course

No, it's a physical reflex, similar in that regard to a man's orgasm, which can be coerced against his wishes.

Are you actually able to back this up? I've known many women who are repulsed by lesbianism and repulsed by the sight of another woman's vagina. Was there a particular study that proves your statement, and if so were the results able to be reproduced consistently?

Falsehoods and confirmation bias annoy the hell out of me. By all means, prove me wrong if you're able, but at this point in time all I am seeing is useless conjecture.

lol

...

All right you fucker then explain how i get extremely aroused by both aras and traps.

Just a quick reminder that Jimmy Saville was never found guilty of anything in a court of law.
If your going to judge guilt by the volume of accusations then basically all famous people are guilty.

MFW I thought you said Arabs,

You're a fucking retard.

It works the same way for kikes.

I'm 100% behind banning tor posters and exiling them to Somalia

American obesity stats are jumped up by obese nonwhites, especially blacks. Pretty much every race except Asians are more likely to be fat than whites in America.

Still too many fat whites though. I blame religion.

(((Who))) could be behind this post? If I were an alphabet agency looking to limit the spread of sensitive information online, and make it easier to trace, I'd shitpost nonstop through TOR to try and get it banned. Don't take the bait.

I've literally never seen a valuable post from TOR here. I get your argument but not sure I agree with it, just based on the reality of what I've seen over the years.

I don't remember women being allowed on this board.
>>>/girltalk/

Please post if you find anything. I'll keep looking.

Since pictures of random normies will include a small number of pedos (1-5%), PU classification might be the way to go. This is a pretty recent paper on it arxiv.org/abs/1605.06955

Basically, as I understand it, Positive-Unlabeled learning is like semi-supervised clustering where you take into account that some of the 'negative' labels are wrong. The IBM Watson guys wrote one using SVM for text.

Oddly enough, nothing asked me if i was female first before i hit "reply". Someone's slipping.

Good point. We can look for ways to improve accuracy later after we have a system running.

"Megan's law" seems to be a gift that could provide us with input data, currently looking into it.

Thanks. Please post findings.

I think if the non-pedo labeled images are going to be random normies, some of whom are or should be sex offenders, then a PU classifier might be necessary.

Breddy gud explanation cs.uic.edu/~liub/NSF/PSC-IIS-0307239.html

I'm still looking for reusable code to do this, but the concept seems simple enough.

Honestly considering becoming an hero, I'm shallow to an extent and don't want to be old and ugly

plug known pedos into deep learning software

This IS what it is, user. Women don't lubricate to just ANYTHING sexual (as the other user implied). They lubricate most in response to the sight of an erect penis, even if they are consciously repulsed by said penis and don't enjoy the sex/rape on any conscious level. It is a reflex left over from the not-that-distant evolutionary past when "erect penis" meant they were going to have a male mate with them and things like "consent" did not exist. Plus the lubrication helps the sperm.

Ignore the other clueless sperging on the topic.

So straight people have fat faces?

kek

Could not find the original link or the study, but here is an article about it:

blogs.psychcentral.com/research/2011/can-you-spot-a-rapist/

women posting demoralizing macro images of my waifu

What's with all these newfags who are so bluepilled on women? Holla Forums lost a lot of nuance once it went mainstream (reddit), and that opens it up to a lot of D&C in areas where it normally could resist such shilling. But all knowledge of women seems to have just disappeared. Even bringing the topic up just brings up shrieks and insults.

sage for off-topic

Are the scared? I start collecting pics of pedos from the public California sex ofender list, to train the detector, and California votes to get rid of the list.

boards.4chan.org/pol/thread/144577083/california-is-removing-90-of-sex-offenders-from

You can download ofender pics from here:
meganslaw.ca.gov/Disclaimer.aspx#

I checked out about 5 states registries, and some are really faggy about protecting sex-offender information. Completely defeats the purpose.

Here's an update on my own progress, so anybody monitoring doesn't get discouraged:

The premise:
The paper by Wang and Kosinski:
psyarxiv.com/hv28a/ [warning: 1/3 of this is virtue signalling to avoid SJW shitstorm]

My understanding of this paper is that they scraped a dating site (probably POF or Tinder) for gay and straight user pictures about 3K for each. They used Face++ to standardize these pictures (Adam Geitgey explains this pretty well here: medium.com/@ageitgey/machine-learning-is-fun-part-4-modern-face-recognition-with-deep-learning-c3cffc121d78 ). Kosinski and Wang then used VGG-face to generate weights for each picture. VGG-face gets about 97% accuracy with the Faces In The Wild dataset, so I think this is a very safe starting place. The Gay-I algorithm reduces the size of the weights (4096 to 500) and then classifies with logistic regression (Maybe a Softmax layer? Logit is a bit vague. These guys are psych researchers, not hard scientists, so I think their understanding of ML algos is fairly superficial). Basically, everything in that paper is completely doable for a normal person because there are available tools that do everything in that process. They cite research suggesting human gaydar is about 60% accurate, so the basic premise of a relationship between facial features and sexual orientation seems sound.

My goal is to follow the same methods. Scrape sex offender registries, normalize the pictures, get weights from VGG-Face, train a classifier with mugshots vs faces in the wild (I'm thinking a supervised classifier to start with [basically because it's only a couple lines in Keras], and a semi-supervised [P-U] classifier with SVM [I would expect this to be more true to reality]).

Challenges:
Sex offender registries are maintained at the state level, and each state is different. I can't write a universal scraper, so each state needs its own. Assuming I could reasonably write and run a scraper for one state in a weekend, I would expect getting every mugshot in the country to take about a year.
Some states give basic victim info, some cite which law was violated, some have tiers for sex crimes. This matters because a guy who raped a 10 year old is a lot different from a negro mating rituals (see: Rape), is a lot different from a guy in his 20s who slept with a 17 year old and a lot different from somebody who got caught pissing in a public parking lot. All could be registered sex offenders. I want to filter out minor offenders.
According to the state of NY, pic-related is white. Basically, demographic data in their registry is garbage.

What I'm doing now:
I have a script that does basically exactly what Face++ does and I have the Torch implementation of VGG-face (I'm trying to get caffe installed so I don't have to fuck with LUA, but caffe doesn't look actively maintained and installation through repos is all fucked). There are several state registries that don't have captchas or other annoying bullshit. I'll start with an easy to scrape registry. I'll save raw pictures in a folder and save names, demographic info, and crime data in a db. That will probably be done this weekend or early next week. I'll take the couple thousand mugshots from that and see if I get any promising results from a trainer. That might take another week or so. If results are good, I'll put it all into a git repo and post here. Then I'll look for a way to get more mugshots to train on.

Do it then faggot or stfu.

wow, that bottom-right picture looks pretty similar to hillary clinton. just more evidence I guess?

...

its not the frozen picture that gives them away.

It's those mannerisms. All those feminine traits and the facial movements. It becomes obvious if you haven't been around faggots in your life.

...

No, it isn't.

They have various civil codes and "I'll kick your fucking teeth in" to avoid it, rather than a Federal system like the USA.

The California registry at the least has a special code for each type of offense. I'm only picking the ones were children were involved (code 288 and some others). What I don't know yet is if to include or exclude the people who got caught with CP in their computer because they seem to be fat losers with hentai.

From the few thousand mugshots I have downloaded, there seems to be clear differences between pedos and rapists. The eyes of the pedos look dead, they look zoned out, and also have crooked lips.

While the registries acknowledge that over 20% of child molesters are women, they keep most of those profiles non-public. From thousands of pictures I have maybe 8 female portraits, and some of them I took from newspapers.

Maybe we should remove the names of the offenders. Less troublesome.

Demographic data is useless, I've read two or three comments from people working on these registries telling that they are told from the higher ups to mislabel on purpose.

I have been looking at algorithms to automatically crop and scale the mugshots, but I'll check what the researchers did in case it is easier to use.

Also, I found this:
news2.onlinenigeria.com/world/565110-controversial-ai-judges-whether-you-are-a-crook-based-on-facial-features.html

I would have opened a repo and upload the pictures I got to test the waters, but what's the best way to do it and remain anonymous?

washingtonpost.com/news/morning-mix/wp/2017/09/12/researchers-use-facial-recognition-tools-to-predict-sexuality-lgbt-groups-arent-happy/

Pretty positive it's possible. Use eigenfaces or neural networks and feed it the correct training data. I.e. pedos and only pesos then test known positive and know negative not in the training set to get acuraccy.

I think their noses gives it away pretty dead on

femanon plz leave.

From my own cursory look, I think there is definitely something about pedos that's consistently off. I'm very optimistic about a decent classifier picking it up. Also, very nice getting that many already.

That's troubling. I wouldn't even doubt it. I wonder if anybody would go on record about it, and, if so, would any journalists publish this info?

Also, check this out: github.com/cmusatyalab/openface

You can use a throwaway email, like mytemp.email to register. I'm not sure what the limits on repo sizes are, though. That might matter when if there are enough pictures. gitlab and gitgud are possible alternatives.

I didn't see that article specify the paper. If I had to guess, it's probably in Chinese. Phrenology used to not even be controversial, though, so I don't doubt there is correlation.

That's basically the idea. VGG-face is a CNN that outputs a 4096 node layer which can be fed into a classifying layer, if I understand it right. I've wasted pretty much all day trying to get a working install of Caffe with no luck, so I'm checking out PyTorch to see if I can use that to get things working. If all else fails, a CNN could be written in Keras, but it wouldn't have the benefit of the proven training VGG already has baked in.

We'll get there, one way or another. It just takes time and a little work.

Going to give pyTorch a chance tomorrow, python makes everything relatively easier.

FYI if you get stuck as here:

datascience.stackexchange.com

Just make sure to ask well thought out questions else youll get ignored.

SITREP:

Some good news, some bad news.

The good:

-CMU's Openface on github has been a huge help, and I've used their code with very small changes for landmarking and normalizing mugshots. Saved me loads of time. Pics related are examples I just made.

- Openface also has a script for feature extraction for a pre-trained torch model (like VGG-face). I expect this to work, but I'm still adapting their code. If that looks bad, then PyTorch should work (and save me from having to learn any LUA)

-I have a working scraper that saves mugshots and inserts data into a SQLite database. I'm just saving all the text related to offenses for now until I can classify and sort it all.

The bad:

-my NVIDIA drivers aren't very stable on my OS. I'm not sure if CUDA is going to behave as expected. If not, my CPU is not going to have enough ass behind it to train a complex model. I may need to rent an AWS cluster with a GPU. Those usually aren't too expensive, though.

-The scraper is gay and based on a selenium webdriver, which is about the slowest way possible that isn't by manual. Unfortunately, curl wasn't a viable option. I also have to babysit the scraper to solve captchas every five minutes. Unless I hit a major snag (like traffic throttling), though, I should have my target state (and about 8K mugshots) after about 48 man-hours.

Overall, I'm not very far behind where I wanted to be. I will only do minimal data cleanup to start with, and, if there looks like there is any promise at all, then I will focus on improving the size and quality of training data. I expect to have a shit-tier classifier maybe as early as the end of this month. My primary plan and contingencies at the moment look like this:
P- VGG-face features are fed into a single-layer sigmoid-activated binary classifier. Accuracy improves with better data. Predictions have enough accuracy for criminal profiling. Pedo empire is crushed. Manking enters a 1,000 year degeneracy-free golden age.
A- DIY convolutional neural net with Keras. Lots of computationally heavy testing required until a working model is found.
C- Semi-supervised SVM model with bias to account for bad labels. This is probably less error prone due to accounting for false negatives, but hardware might be a problem.
E- Naive Bayesian classifier based on facial landmarks and prayer.

If I'm missing something, let me know. I'll make a repository for the model once I start testing.

this warrants a bump

I bet you'd have false positives and false negatives. Consider Stefan Molyneux, he doesn't look Jewish at all (well, maybe all the white admixture redeems him).

Still combining all traits I'm sure it would pick on more than what we do.

Buy captcha solves from deathbycaptcha, if you can program it's about $10 for like 10k solves. I use it for mass account registration.

It will help if you can program*, they have an API you need to call.

Detached earlobes

I could do this, I've used facial landmark software before. Where would I get enough mugshots, with all the jews properly labeled?

usually having to solve a capacha keeps most women out

Correction: someone actually competent could make it for you because you don't know how to create shit.

MJ was a nigger, and he wasn't a normal pedo, he was fucking with White boys for racial revenge

Where is your proof?

The pedo detector already detects jews.

dpscs.state.md.us/sorSearch/

that's MD's public sex crime search. If you were to scrape it for "Charge: Sexual Abuse Of A Minor"
you'll get all the pedos. They are mostly mugshots so you'd get consistent photos. Unfortunately it's MD so you're algorithm would likely be biased towards niggers.

bump

It would just end up detecting blacks and jews.

Lmk about updates. I've been boning up on python for img recog for the last month, would take a look, learning pytorch but not much experience scraping.

fuck this thread
this should be a board
this is just a proof of concept,for /aipol/
how to references would be helpful to mobilizing anons

underrated

I wasn't going to spam this time but then I was banned.

julius classicvacations possibly relieve carol ed accountant undp administrators sleeps. pentagon hai engadget assoc shortcuts brew preface.
portraits engines expectation institution deployed shopping nay col reviewed coping consideration index buried splinter fra shirts snr tbh.

Upload your image sets when you're done, we need to spread them around before they're taken down.

This is another D&C psyop, right? right?

/agree

why use facial profiling when we already have immigration, medical, adoption etc paperwork. Just search for all known permutations of hebrew and weinsteinberg names, and search up and down their genealogical trees however many degrees you need.

just disappear them all out to california around thanksgiving. that's wildfire season, and there are all kinds of big gravel/excavation pits you can just dump them all in with some quick start logs and a couple matches. white liberals won't be able to tell the difference between greasy fire smoke and wildfire smoke anyway.

That explains your anime.

Looks like jews browse this forum
dailymail.co.uk/news/article-3606811/Can-spot-terrorist-just-looking-face-Israeli-company-claims-predict-paedophiles-geniuses-ace-poker-players-analysing-features.html

So here's where I'm at right now:
PyTorch has a load_lua method in its utils class, which lets you read in a .T7 file (which is what you get from here robots.ox.ac.uk/~vgg/software/vgg_face/src/vgg_face_torch.tar.gz). This returns a nn.legacy class, which isn't exactly like normal sequential classes, and there isn't a whole lot of documentation I can find that deal with these specifically. For the last two weekends I've been bashing my head against a wall trying to feed data into the model. It specifies the input dimension as 3 (presumably height, width and color array), but it's throwing indexing errors at me when I send it images (from cv2.imread) or even arrays of images. I'm going to try and make an instance of the Dataset class and see if there is some kind of magic there which the model needs to work. I might also need to transpose the array since I think Torch arrays are backwards by default (color first).

As a fallback, some guy wrote a handy script to take the .T7 model and spit out an ordered dictionary (weights, biases intact). I can't find the link, but here's the script ghostbin.com/paste/kfcsj

So, if I can't make the nn.legacy model work, I could probably rewrite the model in a more layman-friendly framework using the existing weights and just modify the first and last layers. Seriously, though, if somebody knows what I'm fucking up here on the input side, let me know. I'm pretty embarrassed at how much time I've spent trying to accomplish what really should be trivially easy.

If I can't get that to take this next weekend, then I'll try filtering out all but the most serious pedophiles and throw it at SVM and see what happens. Getting 70%+ accuracy on facial recognition is doable with that technique, so I would expect SVM to at least outperform random for this task. From there, it might be possible to generate enough buzz to get more people working on a higher-accuracy CNN.

This is why I think we need a classifier for each race and gender. A racial classifier shouldn't be too hard. I think you can do it with fisherfaces and get 95+% accuracy. I think OpenCV supports this kind of thing. You're right, though. If you throw an unfiltered training set at a classifier, you'll just get black men (which wouldn't technically be wrong, but it isn't what we need)

Here's the scraper I wrote
ghostbin.com/paste/yz95y
I didn't plan for this to be used by others, so there's some things in there you definitely don't need. You will need to re-write the "scrape" method to suit the state registry you're actually pulling pictures from. Also, here is the schema for the "SexOffender.db" Sqlite database:

CREATE TABLE people (id INTEGER NOT NULL PRIMARY KEY, last TEXT, first TEXT, middle TEXT, sex TEXT, race, TEXT, age INTEGER, height REAL, weight INTEGER, hair TEXT, eye TEXT);
CREATE TABLE crime (id INTEGER NOT NULL PRIMARY KEY, category TEXT, v_sex TEXT, v_age_group TEXT);
CREATE TABLE event (pid INTEGER NOT NULL, cid INTEGER NOT NULL, date TEXT);
CREATE TABLE tmp_events (pid INTEGER NOT NULL, description TEXT, code TEXT, date TEXT);

"tmp_events" is me kicking the can on getting a robust, uniform way of classifying crimes (registries tend to give you names of violated statutes, rather than bluntly stating "He blew a 14 year old." Eventually, everything in the tmp_events will be consolidated in the "crime" table, and each offense will be in the "events" table, which is an association table to support JOIN queries (so you can filter by gender, race, offense, etc.). Also the zip code CSV file looks like this:

Zip Code,Place Name,State,State abbv,Region,Latitude,Longitude
501,Holtsville,New York,NY,Suffolk,40.8154,-73.0451
544,Holtsville,New York,NY,Suffolk,40.8154,-73.0451
1001,Agawam,Massachusetts,MA,Hampden,42.0702,-72.6227

It's too big to paste into ghostbin, apparently. I don't remember exactly where I got it, but this is pretty straightforward information to find.

Okay.

My go-to for throwaway emails isn't working right now, but I'll dump the imageset I've got now (5.5K, about 10% are actual pedos) to Mega.

White knights BTFO