Someone tell me why we can't use machine learning to learn what faces go with what tits and start predicting tits based on faces. There has to be a correlation between the two that it could learn and predict.
Someone tell me why we aren't already creating nudes from scratch
Other urls found in this thread:
kvfrans.com
indexxx.com
thenude.eu
en.btdb.to
en.btdb.to
github.com
i.4cdn.org
drive.google.com
twitter.com
Fucken' saved!
If you manage to find a correlation between face and tits you should absolutely publish that as a paper. You’ll get your name on Nature or The Lancet for sure
show bobs and vagene
The you create a NN that can create nudes from any photo.
Damn, just imagine the shitstorm.
>ayy lmao face
>unsymmetrical boobs
>ugly vegana
eww
Sorry for ban.
Yes research GAN but next time use non nude OP pic
I don't see anything in the rules about nudes
not safe for work, enough said.
Check Indexxx or nude.hu both had hunders of thousand models face and nude, tried find some correlation.
What in the fuck is wrong with her left nipple?
>right side of piccy
>left nipple
>lel 69
First you'd have to crop manually thousands of boobies so you can train a model to detect boobie features. Then maybe for the same woman, train some kind of DAN that takes a face as input and tries to match the likeliest boobie.
Not OP or quoted user, but I have a lot of free time, thanks for the tips, I'll start reading about how it works tomorrow morning, cropping shit by night, can't wait to be bullyied to hell for that.
I coud have sworn that I read something about this actually being done.
I was thinking of doing this only with dicks.
pm me your dick & face pics
>DAN
I meant GAN.
Basically if I were to do it, I'd crop the boobs and use some kind of convolutional NN to detect what the features of boobs are. Then because the face features are available by some public methods, I'd associate each face with its boobies, and use a GAN to see how a given face matches to some boobie type.
But this is a lot of effort, you have to use the same picture angle and so on so you can expect your model to work on a hobbyist level.
Also what might be challenging is information on the face likely don't give you much about how large the boobs are. Plus your data is noisy because of fake tiddies.
An easier project might be to associate face with "pink nipple/ brown nipples". This is definitely doable.
is your picture supposed to be attractive or something? she's a gross freckled redhead
I've thought of cropping the boobs, trying to make it recognize the nipples and as much around it as possible, size doesn't really matters, just how skin looks, and then comparing the actual face of the owner of said boobs to the boobs, doing so with like 1k pictures in random kind of woman, I'm pretty sure I can get this much content searching online easily using non nude/nude dedicated sites, using the non nudes as final test, if it can randomly guess the boobs by seeing the clothed picture, it can guess the basic look of boobs by face, after that I can think about what to do with it, probably nothing but if I get to learn something, why not?
complained about this thread an 1.25 hours ago
lil bitch faggit
those tits lmao
I
all this applefags..
not even a real one
No? The freckles and (relatively) pale skin seem to lend authenticity. But either way, she is a qt.
imagine sliding your dick into that.
look at the eyes and eyebrows if the absolutely dogshit artificial dye job doesnt give it away
>ITT: "2/10 would not bang"
you incels never cease to amaze me
>Jow Forums in 2018
Use auto encoder
One network learn correlation face boobs
Other network builds boobs from face
And other check quiality boobs.
Imagine sliding your dick where 30 other dicks have been.
toasty roasty sat on the wall
toasty roasty hit the wall
Why did deepfake porn drop off the face of the earth?
i had a gf with an offset tit due to a botched reduction surgery (why the fuck do women do this)
and it was offputting to be quite honest. i made her keep her shirt on and pulled one tit out at a time.
Begin shitty quality...
You're upset people have standards?
For me the asymmetry of the boobs are not *that* serious but it does knock 1 off her rating, from a solid 7 to a 6.
t. brainlet cuck
>i made her keep her shirt on and pulled one tit out at a time.
uwotm8
>i made her keep her shirt on and pulled one tit out at a time.
You're lucky, my girlfriend had such a large nipple that I could only fuck her in the dark.
>had such a large nipple
>a (singular) large nipple
>had
t. projectionist
There is no way there is a correlation between face and titties. Not size, anyway, you could probably predict colour from the face, but you'd need shape from a top or something.
I'll read it by morning, thanks user, it's 3am and I'm in need of some sleep, but by looking over it, seems like something functional and usable.
WHY SHE NAKED
Wikipedia porn images
Some editors just start take and upload a lot sex/nude content as educative.
> wh*Te women
yikes
no wonder ypipo is dyin out
she still dumped me ._.)
Can't help but imagine all the ass blasted normans
it would probably be even better than the shitstorm of deepfakes
1k is tiny for deep learning standards. If I recall, imagenet has something like 10^6 labelled images.
You can try to use deep features transfer learning but this will require quite a lot of computing power and isn't guarantee to work.
The main difficulty in this case might be of how to extract the features if your pictures have different angles. Different lighting and so on, you can use normalization. But pictures from different angles are challenging with limited datasets and hobbyist hardware.
I was thinking you wanted to train a model that, given a face, reconstructs an "eigen pair of boobs".
If you want to "superimpose" the likeliest boobie pair on top of a clothed pictures, I'd decompose the problem.
Matching shape of body from clothed to naked
Matching boob size from body shape
Matching skin colors
Matching nipple size and colors from facial features.
A good proof of concept might be to start matching face with nipple size and color, as labels and see if you get an acceptable accuracy. If so, then the next step would be to extract boobie features and see what your eigen boobs look like.
After that, use a GAN to match a face real boobs to its generated boobs and see the accuracy.
After that, you're ready to become king of Jow Forums and have your name featured on every leftist tech blog .
>an hour ago
are there even mods on Jow Forums anymore
it can and I am
the problem is not the NN but the collecting of 200,000 dressed and naked pictures
>adds pic to dataset
She has a cute face and a nice body, and her pussy looks great. Her tits are a little retarded but that's a good thing because she'll lower her standards so far that even you might have a chance.
You're getting there user!
yikes
Metart,angels amours,hegre,nude in public,voyeur sites,use torrent easy build Terrabytes nude photos or videos.
Begin amazing today begin build monumental semi public nude young attractive women dataset.
so hot thx user i came on cat and he hiss at penis
once I get on RTX I'm taking this really serious again
I will come here for the announcement
Isn't a good way to gather data just making a bot sit on (3D) porn boards and scrape all images from boobs related threads?
Something were anons contribute with their porn would be interesting
You're a faggot if you wouldn't fuck her. Also a virgin if you think her pussy is ugly.
Doing god's work user
There's gonna be plenty of thot photos to work with
> immediately resorts to virgin shaming and homophobia
yikes
any way we can help you?
The future looks bright
>using yikes ironically or unironically
> censoring yourself in the filename
yikes
so᠌yboy cuck confirmed
For example could select per model
indexxx.com
Per photographer
thenude.eu
Get torrents
en.btdb.to
Even torrents for site
en.btdb.to
Terrabytes high quality photos.
New version tensorflow and pytouch, new version cuda 10 and support for tensor cores.
>spamming "yikes"
It's true though, literally nothing wrong with that cunt.
the hardest problem with the data is finding pure pictures. when you go scraping you run into a lot of half naked girls and I autistically go through all these. I did write a script that cuts the images in half from reddit/dressed_undressed
Thanks. Will check
I could personally help you find more pictures if you need help, plus I'd be interested in your code and maybe help you with it
I I've seen your posts on Jow Forums, maybe 6 months ago when the makeup app thing happened.
Reddit just recycle old porn websites or porn forums.
Some porn products take very uniformal dress/nude photos for casting, even begin some kind fetishe, image f a p had a lot sets like these.
Google
“czech dress undress”
Begin several companies, then search for torrent entery photo sets
>2.5 hours in
>OP pic still up
Nothing right with it either
>he doesn't like thick labia that grip your dick and provide suction
Faggot
True, it's not on my dick right now.
uhhhhm sweaty, this is an educational image. get your porn obsessed mind out of here
>tfw candid nudity is more arousing than deliberate porn
We /b/ now, sadly
AHEM! While you focus on bobs and vagenes this gentlemen does the real job.
github.com
Kids will end up combining it w/ face aging software to look at their own or each others' made up tits. Every young boy will end up fapping to his own tits.
fuck off fag go back to or something
>The decensorship is intended to ONLY work on color hentai images that have minor bar censorship of the penis or vagina.
>It does NOT work with:
> Black and white images
> Monochrome images
> Hentai containing screentones (e.g. printed hentai)
> Real life porn
> Mosaic censorship
> Censorship of nipples
> Censorship of anus
> Animated gifs/videos
>In particular, if a vagina or penis is completely censored out, inpainting will be ineffective.
it's fucking nothing, you could probably write a photoshop script with content aware filling that would do a better job.
Have you seen /b/ lately? It's such a shame.
I made it for the technican challenge, so as soon as I saw it would actually kinda work I put it on hold. they took 20 days to train (I'm still on kepler)
as part of the technical challenge I did make apk's already
if you guys can keep this thread alive I'll try to find the apk's I made
where can I upload these?
do keep in mind these images were cherrypicked and it really isn't that great (yet)
Literally photoshops content aware fill that's been in photoshop since 2010.
Mega, mostly thread will dead because OP imagen.
You had amazing lucky, millons photos from Instagram women in bikini and easy to scrap.
Nvidia builds massive fast for training or inference for new Turing and popular frameworks adopt in some months.
based and redpilled
i can make the logo
we can also make a new machine tits general user
>dem nails
into the trash she goes
End goal is made Honey select but photorealistic.
Build virtual humans and having advance editor for skin,nipplers,boobs,face,eyes,hair and pussy.
I too am interested in this topic.
I'd be more interested in the network structure you used.
Did you just took an existing image GAN and with cropped nude pics as training data?
If she had a cock she'd be perfect.
oh shit, I'm happy I'm not currently in the work place. I sometimes browse Jow Forums there
it's the app called "tf stylize"
I always thought there might be a correlation between the way you sound and the way you look