Jump to content
The Corroboree
Sign in to follow this  
Yeti101

Inceptionism: Trippy output from Googles neural neworks

Recommended Posts

wow. that's nuts.

Share this post


Link to post
Share on other sites

such a huge territory i think they're entering, by applying the power of computing to the vast arena of all things visual. i don't see why descriptions can't be added so that dogish things or duckish things can either be slowish or fastish, standy or sitty, and so on.

allowing random backgrounds to become part of its lexicon seems like it would occasionally help with ID but usually hinder it.

never heard of an artificial neural network before.

Share this post


Link to post
Share on other sites

Kinda like the webcam on its own image on overdrive.
Gotta love feedback :P

Share this post


Link to post
Share on other sites

As the person who put me on to this said:

"We monitor many frequencies. We listen always. Came a voice, out of the babel of tongues,
speaking to us. It played us a mighty dub."

"Call 'em Winter Mute" said the other, making it two words.

Interesting indeed. I wonder what it's like to be that neural network.

Share this post


Link to post
Share on other sites

maybe your visual cortex could give you a little insight? i mean, maybe i'm jumping the gun, but i clicked on the gallery these time and see a common theme which is familiar to me. that melting through time rainbowish fractal dealio.

Share this post


Link to post
Share on other sites

I agree, there is something very familiar about some of the images.

  • Like 1

Share this post


Link to post
Share on other sites

here's a video tut on the kind of thing this is using (convolutional neural net) if you're curious

it's about a simple as you can make it

  • Like 1

Share this post


Link to post
Share on other sites

in the first 40 seconds it sounds like he keeps saying LSD then "and this stuff that alex grey has been using for a long time" bahaha

Share this post


Link to post
Share on other sites

Couple of updates:
The source code for the program that makes these images has been released: http://googleresearch.blogspot.com.au/2015/07/deepdream-code-example-for-visualizing.html

Motherboard (that veritable fountain of quality journalism) shows what happens when you put porn through this process - NSFW!: http://motherboard.vice.com/en_au/read/what-computers-dream-of-when-they-look-at-porn-nsfw

(Machine learning side note: auto-correct keeps trying to change 'porn' to 'poon'. How is that better?)

The 'Deepdream' sub-reddit has heaps more info and images https://www.reddit.com/r/deepdream/

  • Like 2

Share this post


Link to post
Share on other sites

.delete.

Edited by ThunderIdeal

Share this post


Link to post
Share on other sites

the process is different to the one described earlier where an image of pure noise is optimised to look more like the software's learned concept of 'banana'.

these seem to take a pornographic image and optimise it to look more like a dogs face or a crustacean.

the kind of process described towards the bottom of this link http://googleresearch.blogspot.com.au/2015/06/inceptionism-going-deeper-into-neural.html

porn isn't exactly the most interesting use of this process but that seems more interesting to me, letting it learn what porn looks like and asking it to pornographise a nice sunset photo.

i would like to figure out whether similar processes are going on when a person uses drugs, or if it's just some of the results that are similar.

Edited by ThunderIdeal

Share this post


Link to post
Share on other sites

Not sure Thunder, I'm still getting up to speed with all this.

Share this post


Link to post
Share on other sites

i really like the selfie mirror one, wether that was human created or by a program, that really is awesome art

  • Like 1

Share this post


Link to post
Share on other sites

both of you ^ i modified my above posts after i realised what's going on

it's really interesting how computer technology has paved the way to discoveries which have so much in common with the subjective experience of tripping.

did people trip the same way before computers?

it goes right back to the advent of chaos theory and the creation of complex fractals both of which relied on computer power.

if you read the last bit of my last link where it talks about zooming each iteration, you're starting to enter territory maybe somewhere between pure math fractals and the fractal geometry of messy natural things.

Edited by ThunderIdeal

Share this post


Link to post
Share on other sites

I'm glad someone knows what's going on :)

Seriously though, if windows hadn't just eaten itself on my PC, I'd be installing that stuff already.

Share this post


Link to post
Share on other sites

Need to put your pic through it Hillbilly.

Share this post


Link to post
Share on other sites

ask that question while you're tripping, then report back with answers even more questions

  • Like 2

Share this post


Link to post
Share on other sites

Don't head down that rabbit-hole my goaty friend, that way lies madness.

Share this post


Link to post
Share on other sites

i would like to figure out whether similar processes are going on when a person uses drugs, or if it's just some of the results that are similar.

Well I guess there's something similar going on, only our pattern-recognition isn't really iterative (afaik anyway?). But in both cases you're taking a pattern-finding machine & giving it something to look at, then removing some barrier to it recognising more patterns - in the case of Deepskywintermutenet (or whatever its called) you're telling it to keep looping until it sees more dogs* , in the case of a tripper you seem to be removing some of that post-production work, so to speak - all that filtering & processing that goes on before you decide what an input actually is** Because reality is some trippy shit, it takes alot of work to make it look that normal. Your brain is constantly having to do stuff like tell you "look I know your skin looks green right now but that's just the lighting in here ok, I promise it's still skin-coloured so don't start freaking out on me - remember how your skin looked in the other room, well it still looks like that", or "wow, clouds are REALLY fractal, I mean there's like an infinite amount of visual information in there, so I'm just gonna tone things down a little until they just look kinda grey & blobby so you're able to walk along without running into stuff". And when you're tripping, alot of that gets shut down. Our brains are really good at finding patterns - often too good, and so we end up with racism & astrology & all kinds of other faulty logic & outright mental illness. But those same processes that are in play when your aunt decides that all women called Tina are cheating sluts, or that Scorpios are queue-jumpers or whatever, don't just apply to people she doesn't like - that same processing happens to everything. It takes constant effort not to see patterns everywhere - our brains really are wired for that. So when you add that innate tendency to some drugs that help it along, and then combine it with the fact that reality actually IS full of rainbows & fractals, yeah I can see how you end up with images like those above. Computers take alot of training to see patterns, it doesn't come naturally to them, but once they've got the idea there's no reason that their pattern-finding tendencies couldn't be turned up or down via programming in some way that's analogous to us taking drugs. I guess you could think of the progressive iterations which find more and more dogs as being something like an increasing dosage.

*(actually they seem to be telling it to keep looping until we can see the patterns - I wonder what the results would be if you let the program itself be the judge of whether it had altered an image enough to be recognisable as ___ ...ah, trippy shit indeed!)

** I don't think there's much real evidence for this theory, I mostly came up with it by reading a little neuropsychology & a lot of sci-fi

  • Like 1

Share this post


Link to post
Share on other sites

You know what I was saying about how we apply the same pattern-finding processes apply to different things, well here they got neuralskynet to use the same process on sound files...

The results aren't as cool as the visual ones yet, but on the other hand they haven't dedicated anywhere near as much effort to training the program on audio, and wow the potential!

  • Like 2

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this  

×