Hacker News new | past | comments | ask | show | jobs | submit login
This Cat Does Not Exist (thiscatdoesnotexist.com)
428 points by ChrisArchitect on Feb 20, 2019 | hide | past | favorite | 180 comments

I think this might be a well-done satire showcasing the pitfalls of having insufficient training data. It's magnificent. Half the cats are adorable and half are, I believe, creatures from "The Thing".

> pitfalls of having insufficient training data

Also when you use a dataset containing watermarked stock photos and the watermark is treated as a signal instead of noise :) https://i.imgur.com/O1ZRwSF.jpg

They look like Star Trek transporter accidents.

That could mean a lot of things. Are you saying that they look dead, fused with another cat, drastically younger, attacked by an interdimensional monster, cloned or are the evil alternate reality version of themselves?

Fused with a human in some cases, so very much like "The Fly" sometimes: https://imgur.com/a/usixtLS


All of the above.

Reminds me of Alien Resurrection. The other Ripleys.

"She's suffering horribly, my god. [Weeps] I will mercifully end her suffering .... with FIRE." ?WTF

Or "The Fly" missed experiments.

Complete nightmare fuel.


The cats aren't so bad. It's the "people".


thats the francis bacon painter plugin

Knowledge is power, France is Bacon

When I saw the first Google deepdream images, they made me physically nauseous.

But now, in 2019, I've fully acclimated such nightmare-fuel concoctions: "Oh look, another GAN cat!"

I raise you again: https://imgur.com/a/Dc9uyUY

The Ring, Cat Edition.

Tomorrow's Google reCaptcha

I suspect that the problem is less one of insufficient training data, and more one of excessively noisy training data.

There are three of these generators that have shown up on HN in the last few days—people, cats, and anime faces—and what the other two (more successful) ones have in common is that the things they're trying to generate all have the same basic shape and structure: that of a face.

The cat images in the training data are clearly just cats from every angle. There's much less of a clear structure for the neural net to recognize and reproduce.

That it's doing as well as it is is actually kind of remarkable, given that issue.

... and thank God it doesn't. Just look at this poor fella!


Did it automatically generate that "caption" because it learns from lolcats/cat memes? Fascinating!

This reminds me of the movie Annihilation, it’s like it’s blindly mimicking something without knowing what it’s doing.

So weird!

> it’s like it’s blindly mimicking something without knowing what it’s doing

If it's not, that should be the definition for machine learning.

I also got a few characters of would-be-text into a couple of images. As you said, the net was probably trained with images that included cat memes and it "learned" that meme-text means "cat".

Makes me wonder if it is possible for deep learning to create an artificial language. Supposedly Lojban's vocabulary was created by an algorithm, although how it got "mlatu" for cat is obscure.

Here's how Lojban's vocabulary was created: https://lojban.github.io/cll/4/14/

> At least one word was found in each of the six source languages (Chinese, English, Hindi, Spanish, Russian, Arabic) corresponding to the proposed gismu. This word was rendered into Lojban phonetics rather liberally: consonant clusters consisting of a stop and the corresponding fricative were simplified to just the fricative (“tc” became “c”, “dj” became “j”) and non-Lojban vowels were mapped onto Lojban ones. Furthermore, morphological endings were dropped. The same mapping rules were applied to all six languages for the sake of consistency. > ...

Never heard of Lojban's vocabulary - thank you for a very interesting read!

Is it invoking a forgotten elderitch prayer to Chthulhu?

Cute idea, but about half of these have given me very weird/unrealistic results. For instance this[1] one which, while amusing as hell, isn't exactly convincing.


Fun fact, the toroidal cat is actually a stable gravitational configuration along with a spheroid. There are upper and lower bounds on angular momentum and mass of course, but within those bounds you can have a toroidal cat orbiting its primary!

Wait, what’s a cat again?

Complete tangent, but I’ve always wanted to ask someone else who thinks about toroidal gravitational physics: in the center of a toroidal cat (planet) of sufficient mass, does gravity “cancel out”, or are you instead being “pulled apart” in every direction at once, as if on a medieval rack?


At the very center of the hole, everything cancels. But it is unstable. As you head in any direction, you will get pulled further. Which means that there are also tides pulling you apart.

At the center of the solid ring of the torus, everything also cancels. But this time the tides are squishing you together. Which is why the configuration is stable.

Cancels out. To pull something apart, there would have to be a gravitational gradient, meaning the object would have to be pretty huge relative to the toroid.

isn't that the same type of cat physicists used to proof the buttered-cat paradox ?

Actually the buttered cat hypothesis has been proven with much less exotic cat shapes. I recall seeing a paper showing that a cat with dimension n=1 and zero rotational inertia would spontaneously begin spinning as a result of the presence of a "butter field." Of course the experimenters ran out of point cats and had to go to the store to get more syrup shortly after completing the experiment, so it has been rather difficult to replicate. However the authors assured readers they would attempt replication at the next breakfast fundraiser for the department.

Thank you for an opportunity to revive this video - https://www.youtube.com/watch?v=Z8yW5cyXXRc

I wish researchers were using these examples to further understand what this network does, how it fails and what are its fundamental limitations. However, such digging would undermine the hype, so I'm not particularly hopeful. Most of the issues are just written off as kinks to be ironed out.

Another thing that really bothers me is that no one tries to replicate any of these results without neural networks[1]. To most people here this is the natural result of deep neural networks being the bestest algorithm ever. To me, this indicates that much of the current ML research fails to generate true insight.


[1] For example, what would GAN-like architecture look with gcForests? No one seems to care about questions like this, even though gcForests have tons of practical advantages over neural nets.

"To me, this indicates that much of the current ML research fails to generate true insight."

I don't think anyone who knows what they're talking about would say otherwise.

ML isn't perfect, in case you didn't know. If you want to catch up with academic progress, search for stylegan. Aren't we allowed to have some fun sometimes?

To be fair, the site isn't called "This is a realistic looking cat the doesn't exist"

I got a cat that could easily fit in MoMA.


Author may want to implement a labelling method for users for a days to maybe train the discriminator a little bit better. Would be a cool human-in-the-loop exercise.

Yes something like this: http://www.whichfaceisreal.com/

I was surprised to see that the easiest way to figure out if a face was real was by looking at the background. The face generator seems to be terrible at everything but faces. There are often strange visual artifacts and clipping issues, and the face generator never seems to put another person in the background of the picture.

Wow. I got 2/10 the first time and then did again after reading your comment and got 10/10. The background really does give it away.

also, the teeth. in generated images they are usually strangely asymmetrical

I got some pretty tough ones on mobile. Maybe if I had zoomed in I would have seen the differences.’, but eh: https://imgur.com/a/UXSCOG8

Very unusual expressions seem to me likely to be real, simply because the generated images are AFAIK in a sense statistical averages.

I thought this too! Then I picked a weird expression and was wrong lol.

The fake image looks extra weird because it looks like he has glasses on the right side of his face but its kinda cropped out

Seems like the algorithm considered his baggy eye pockets as the bottom rim of a pair of eye glasses.

Ahah. I didn’t see that, and I have pretty bad vision myself.

I read this as "which France is real" and was slightly disappointed when I wasn't able to test my incomplete knowledge of European geography against a neural net.

ML generates some rather bad artifacts. Just look for those.

Even in this[1] difficult comparison you can see the non-human repeating skin patterns on the right and the awkward teeth contour. Also hair-on-skin often looks wet and with unnatural bends.

When comparing wrinkly people then it gets a little harder.

1: https://i.imgur.com/KWnhkBS.jpg

That one is super hard when looking only at the face.

Look at the clothes and necklace. The clothes are different on left and right sides of her face - the moment you see it you can't unsee it and it's obviously wrong.

After yesterday's 10 minutes of watching those fake faces this test was super simple for me, I did like 25 without mistakes which kinda shows the fake generation has a long way to know to fool good eyes.

Look at the eyeball, compare reflection, it’s easy this way.

Wow. I have no idea which face is real.

You can pretty reliably guess correctly if you look for ghosting/blurring/chromatic aberration along sharp edges, e.g. around the eyes, on the chin, and in hair. ML hasn't quite mastered the fine details yet

I have found this: https://imgur.com/a/NgcEXVF

Corgi breeders don't stop to think if they should have.

how one can see Donut Cat and suggest the algorithm be changed to snuff out his non-existence-existence, is beyond me

looks like mighty fine kittens to me : https://i.imgur.com/wbZmNzY.jpg && https://i.imgur.com/rmxEkdC.jpg

Interesting. I looked at about 20 of these and only saw 3 that didn't look like cats. I guess I got lucky! Now to hit refresh some more times to find some weird and wonderful mutant cats.

wow - a rare breed - the donut cat!

Oh, the elusive myhological doughnut cat. Finally confirmed!

Most of the ones with humans don’t turn out so good.

Someone download a bunch of corporate ppt slides from slideshare and train a model suitable for thisslidedoesnotexist.com.

I think we've seen what these latest gen GANs can do with natural images, so why not try a novel application ?

They can do something to imitate text in images, so there's reason to think ppt slides should work.

I was about to comment that this sort of thing would be a great source of "stock" pictures for powerpoint presentations; pick a theme (subject) and hand-pick the good ones.

Also this airbnb doesnt exist https://thisairbnbdoesnotexist.com/

and these recordings dont really exist https://www.youtube.com/watch?v=llG-jQf8IBk&feature=youtu.be

And these Stack Overflow/Exchange questions don’t exist:


Disclaimer: Fun useless side project by me (And contains no machine learning, but just plain Markov chains)

you have no idea how similar this is to SO's moderation queue content

I'm really exited for https://thiswebsitedoesnotexist.com

You don't need to wait. That site does not exist until someone books thiswebsitedoesnotexist.com domain.

  Creation Date: 2005-08-26-T19:04:47Z
  Updated Date: 2018-08-23-T01:12:14Z
  Registrar Registration Expiration Date: 2019-08-26
And I'm pleased to say that (as I write this, at least) the registrant has done exactly what I had intended to do with it.

Looks like there were some memes with watermarks in the training data https://i.imgur.com/iMOSUog.jpg

That's a caption, not a watermark. This one has a watermark, and was clearly taken from that popular vendor of stock images, shuttersrstsck:


Raises some questions about what is able to be copyrighted vs. derived works if the generated image was produced by this algorithm and doesn't actually exist in Shutterstock's (excuse me, Shuttersrstsck's) database.

I was referring to the gray bar with text on the bottom. Its supposed to have the URL for the website rehosting image, like the bottom white bar in the image you linked to.

Some of these are the stuff of nightmares.

Yes, that's one of Cthulhu's favourite memes.

Half, if not more, are nightmare fuel.

I didn't take me long to get this abomination https://imgur.com/a/gI8Vlla

There's another one too: http://thesecatsdonotexist.com/

You're absolutely sure none of these exist? This look way too real: https://imgur.com/a/aQYjeL6

may be it was supposed to be http://thesecatsdonotexistANYMORE.com/

funny take on This Person Does Not Exist and StyleGAN generation (related post: https://news.ycombinator.com/item?id=19144280)

I still can't shake the feeling that most of these StyleGAN images are cleverly overfitting and just showing the face of an already existing cat in its training data. (But would love to be proven wrong!)

Back when I used to experiment with Markov chat simulators, this was a big problem. Besides the disappointment of finding out a particularly clever generated sentence was actually verbatim from the training set, there's also "accidental sharing" and/or "plagarism" issues. Of course with text it's pretty simple to code a check that output doesn't exactly match any known inputs. Not sure how you'd do that with images; maybe some kind of image hashing. (I wonder if you could use the neural network itself to assist in that - i.e. hash the measured values at a lower-dimensioned layer of the network rather than the raw image.)

Yes. For hysterical raisins, VGG-16 is usually used as the hash/space for the nearest-neighbor lookups. Recent example of this is in the BigGAN appendix, where you can see that despite the dog samples looking perfect, they are nevertheless totally different from their closest neighbors in the ImageNet training data and so can't be memorization.

The website owner could use the time between two refresh by a given user to assert the "weirdness" of a picture.

Curious, is cat more complex object than human's face for ML to be trained and produced?

The human face training data was probably much more uniform: all well-lit photos of human faces in the center of the photo looking directly at the camera. Whereas the training set for this was probably any photos of cats, in any lighting conditions, with the cat in any pose.

It looks to me that cats as a whole are more complex than human faces. I suspect we would get similar results with full humans. Cats have huge variation in color and pose that faces do not. An additional factor is that most portraits have the background in a separate focal plane while most cats are photographed against a complex background.

Might be that there is a difference between the quality and quantity of the training sets.

Might be that humans are not very good at recognizing differences between cats.

It would be interesting to see how realistic a cat thinks these are, maybe by measuring brain activity or reactions. It's possible that a cat may not be fooled by cats we think look real, or perhaps more interestingly, that a cat is fooled by a not particularly good image.

Common cuckoos lay their eggs in other birds' nests. The chicks don't necessarily look much like the host species to the human eye, but they can fool their hosts along the correct dimensions to get food from them. It's an interesting question to what degree ML algorithms trained on human dimensions could be foiled by an animal whose brain has been wired for different perceptions, or how feasible it is to train an ML algorithm on animal perception, or if it's possible to make an algorithm that successfully fools, say, both man and dog.

On the last point, for example: to make fake sounds that fool animals with different hearing ranges, presumably you have to be able to output sounds across the union of the ranges and train on sound data over the union of the ranges.

(Note: I'm not a biologist, if someone more informed wants to correct me on anything here you are welcome to do so.)

Starting a lab soon to investigate this. Brb, I'm going to gather feral cats.

In a plenty of results, I'm getting an ok cat face with a cat-like blob attached to it. So I'd say it's difficult for the model to discern any features in a mass of color-splotched fur.

Two creatures in this picture: http://imgur.com/TDeLYwF

One looks like a cat, the other...?

Select the pistol, and then, select your cat.


Can someone explain to me what I am looking at? :)

Someone recently posted https://thispersondoesnotexist.com/ -- which shows a random AI-generated human face.

This is a cat version, but it works out a lot worse. Some are fine, some are terrible.

I think that's on purpose - the training data appears to be all results from search for "cat pics"... captions and meme text and all.

kinda funny how this GAN picked up the strong presence of cats within early 2010's memes, some of the resulting photos have remnants of the distinctive white-on-black text from some of the training data

Being reincarnated in a cat isn't always great. https://imgur.com/rHm8Nwv

I'm glad this doesn't exist. https://i.imgur.com/xrhRpAe.jpg

Don't they? I feel like half the pictures are probably nearly identical to some picture in the training data and the other half are distorted mess...

You bet it doesn't...


Hey! It's my little Tiddles the cat!

Human-cat hybrids: https://imgur.com/a/mqBA8IM

This generates nightmare fuel: https://imgur.com/jw9OmFi

Please do not change the training data for this....or at least keep this version. These are the most amazing failure cases ever.

If you want to see this more often, I made this Chrome Extension for fun: https://chrome.google.com/webstore/detail/this-cat-does-not-...

A while back I started to think about something like gan.tv (it's an owned domain I see) for computer generated entertainment. This cat one and the person one would be example channels. I assume we're going to get pretty creative with how we use automated creativity in the near future.

Looks like taxidermy'd cats, lol.


for some reason shutterstock showed up on the bottom of the preview image when i posted it to my discord channel

For the sake of some of these cats I am happy that they don't exist.


for some reason shutterstock showed up at the bottom of the discord image preview XD

On one reload, I got Grumpy Cat. The background was different from any image I can find in a quick image search, but the cat was definitely Grumpy Cat. Does that make the domain name a lie?

I also had Grump Cat. Looks like their dataset was not large enough to avoid overfitting.

meanwhile: This Pidgeon DOES exist https://en.wikipedia.org/wiki/Pigmy_Pouter

Yikes I hope not. What is this eldritch abomination?


Well, at least after the human race is wiped out by AI, the fascination with cats will still live on in the new sentient creatures dominating the planet.

First person to create "thisdoesnotexistdoesnotexist.com" with some appropriate functionality wins one Internet. And apparently several HNs.

Dear god what have we done https://i.imgur.com/OanNO7W.jpg

I.. don't get it. I imagine it's embarrassingly straightforward once I do, but until then I'm lost. Could someone help me out?

The missing piece for your brain algorithm must be: https://thispersondoesnotexist.com/

Jesus christ I sure hope it doesn't


I'm so sad I didn't screenshot the first image with "Shutterstock" written all over it - this never happened again.

If it makes you feel any better, I've seen with with a watermark too!

http://notarealhuman.com/ has cats as well :)

I love this, but it's a lot less uncanny than thispersondoesnotexist.com due to the number of vaguely cat-like blobs it generates.

There are frequent bugs where the cat face is at a weird place, so I'm glad nature doesn't let such animals actually live.

https://ibb.co/x55vHwv Another failure sample

The domain can as well be petsematary.com

I'm really tempted to register http://petsematary.ai and link it to this, but it's a bit pricey for a prank.

FTI: "Pet Sematary"[sic] is the name of a book by Stephen King: https://en.wikipedia.org/wiki/Pet_Sematary

Saw a cat with a bunch of garbled motivational poster text. Drove me over the wall with laughter at work.


Failure sample

at least not on this plane of reality it doesn’t, maybe one of the inner circles of hell tho (yikes!)

I'm curious now what would happen if you trained a model on 50% cat pics and 50% human pics

I clicked on the post expecting an article or a Github repo, but I was surprised to find a cat.

This site made my morning. Wow such incredible technology! Machines are learning too much

So are these from real cat photos.. or they are entirely made by AI? I am confused...

Im curious what is generating the pictures? I want to play around with this toolset.

It's only a matter of time before thisporndoesnotexist.com happens.

Holy Shit! I hope those DEFORMED cats don’t exist lest we put them down!

Great idea, poorly executed

Should be called "this cat was in transporter accident"

I wonder how well this'd work on screenshots of websites.

This technology would be pretty cool for placeholder images.

I'm waiting for "this alien does not exist".

This actually looks a lot like failed taxidermy

The first five I reloaded were all cross-eyed.

Thank god! some of those were monsteras.

I have nightmares about these cats.

A reload page button would be nice..

Now this is getting out of hand!

I found some horrific misses...

And if, it's on lsd.

Well I should hope not!

Schrödinger’s cat?

Looks at cat

Sees title


ceci n'est pas une cat

Coming next thispoodoesnotexist.com

Do you really want to gather the training data for this one?

That'd be a bit of a poop job.

Quick, someone make a slack plugin for this!

I certainly hope not, the first "cat" I was shown had a human face.

ThisCarDoesNotExist or ThisDressDoesNotExist would be fun. Can see them monetized nicely.

Ack. Almost every other made me say “I should hope not!”. A lot of deformed kitties there, algorithm needs work.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact