|
Syntopia
|
 |
« Reply #75 on: July 08, 2015, 11:39:28 AM » |
|
But the most disturbing results are probably using food:
|
|
|
|
Logged
|
|
|
|
|
Chillheimer
|
|
« Reply #76 on: July 08, 2015, 11:54:47 AM » |
|
As soon as anyone knows of a new tool or an (relatively) easy way to set this up on a windows machine, please post here!! the new zainshah site doesn't allow multiple tabs and is far slower then the first...  |
|
|
|
|
Logged
|
--- Fractals - add some Chaos to your life and put the world in order. ---
|
|
|
|
kram1032
|
|
« Reply #77 on: July 08, 2015, 04:43:18 PM » |
|
I love the food pics and actually, somehow, the blindness test is one of the most interesting images I've seen thus far. I'd love to see a higher resolution version of it to see what patterns emerge within circles. - Although one ought to be careful with that. Eventually the results will mostly look like those of all flat areas, essentially random, only changed up by the border regions. It seems to me that the network has some kind of "preferred size" at which it is most likely to understand (or not) its input.
|
|
|
|
|
Logged
|
|
|
|
|
kram1032
|
|
« Reply #78 on: July 08, 2015, 05:01:16 PM » |
|
Another video explaining why this work was even done, as well as showing off other visualization techniques.
<![CDATA[<a href="https://www.youtube.com/v/AgkfIQ4IGaM&rel=1&fs=1&hd=1" target="_blank">https://www.youtube.com/v/AgkfIQ4IGaM&rel=1&fs=1&hd=1</a>]]> |
|
|
|
|
Logged
|
|
|
|
|
cKleinhuis
|
|
« Reply #79 on: July 08, 2015, 05:09:55 PM » |
|
very cool, and what this whole thing shows is that neuronal networks actually work, so, contrary to computer science which poses a limit to mechanical machines, it is not needed to reach ultimate non-determinism its just enough to provide all the kind of informations, and neural networks provide a method to actually handle every input, so in fact this approach to make a machine actually talk, or think seems to be really just a step ahead, on this step translations and various other stuff can be done by them
the thing that scares me is that we in fact render us people obsolete, lol
|
|
|
|
|
Logged
|
---
divide and conquer - iterate and rule - chaos is No random!
|
|
|
|
Fitz
|
|
« Reply #80 on: July 08, 2015, 05:57:29 PM » |
|
the thing that scares me is that we in fact render us people obsolete, lol
I personally look forward to the dog/pagoda-based future the robots have in store for us. |
|
|
|
|
Logged
|
|
|
|
|
kram1032
|
|
« Reply #81 on: July 08, 2015, 06:16:23 PM » |
|
with all this training on dogs they'll think dogs are their original overlords and humans mere pets
|
|
|
|
|
Logged
|
|
|
|
|
cKleinhuis
|
|
« Reply #82 on: July 08, 2015, 06:56:41 PM » |
|
naah, you all know the systems are far to small to overcome the actual problem, the EU is developing a neural network of the size of a human brain ... image recognition is a sub part of a larger "brain" much more components, like audio recognition, speach recognition, smells recognition etc have to be trained as well to have a base
|
|
|
|
|
Logged
|
---
divide and conquer - iterate and rule - chaos is No random!
|
|
|
|
kram1032
|
|
« Reply #83 on: July 08, 2015, 07:09:55 PM » |
|
Smell recognition, huh? Well that ought to take some time. Smell is the sense we know about the least and also the one we can replicate the least well with tech, of all our senses. (And if it becomes easily reproducible, I certainly don't want the inevitable addition of smellovision  ) |
|
|
|
|
Logged
|
|
|
|
|
Chillheimer
|
|
« Reply #84 on: July 08, 2015, 08:19:30 PM » |
|
I did it! !!!!! whuaaaaahhhhhh!    I got it set up! my workstation is dreaming right now!!!!! phew. took my the whole afternoon... It's coommandline stuff and I have no idea what I am doing and where all that 4gb of stuff installed itself to, but its working! not comfortable at all. but does the job!  I don't know the last steps exactly and what lead to it finally working, but if you try this guys version, it obviosly is manageable for a non-coder like me: https://github.com/Dhar/image-dreamertry at your own risk!! and feel free to ask me, maybe I recall what I did when I was stuck at the same point - probably not..  ---- wow. this is slow! and it doesn't solve my problem of how to do larger resolutions than 1024*1024. Seems like memory kills it. damn. doesn't make use of my 16gb ram. why not?! aww.. all this for nothing?! well, at least I can continue to dream low res without the website.. |
|
|
|
|
Logged
|
--- Fractals - add some Chaos to your life and put the world in order. ---
|
|
|
|
Chillheimer
|
|
« Reply #85 on: July 08, 2015, 11:06:07 PM » |
|
We give also tribute, to this artist, Memo Aken, aka Chillheimer,
that made this first iconic film of the new fractal style, so rich, so mesmerizing, from its roots to the outer limits, entitled," Journey through the layers of the mind"
I really wish I were that guy and could do what he did - but I'm not.. I didn't intend to make that impression..  has anyone tried feed back? like taking the above image and sending it through another dream cycle? of course, i did this with all of my pics, sometimes 5 or 6 times and changing between the 2 different algorithms.. |
|
|
|
|
Logged
|
--- Fractals - add some Chaos to your life and put the world in order. ---
|
|
|
|
|
|
3dickulus
|
|
« Reply #87 on: July 09, 2015, 02:26:12 AM » |
|
wow, that's disturbing, but...
<![CDATA[<a href="https://www.youtube.com/v/X_tvm6Eoa3g&rel=1&fs=1&hd=1" target="_blank">https://www.youtube.com/v/X_tvm6Eoa3g&rel=1&fs=1&hd=1</a>]]>... so is this  chatbots? (I know a bit off topic but could be the audio interaction portion) |
|
|
|
|
Logged
|
|
|
|
|
3dickulus
|
|
« Reply #88 on: July 09, 2015, 04:23:15 AM » |
|
Color blindness tests proved surprisingly robust to inceptionism. Here is an example, but I had to crank up the step size:
Really interesting! when a human takes that test if we recognize the numbers or letters then we are not colour blind, I suppose that another test with a grey-scale version of that image that has no discernible pattern would yield a similar result, if so does this mean that NN is colour blind? given that it's only trained with dogs or buildings and not text or numbers I guess this result could be expected. |
|
|
|
|
Logged
|
|
|
|
|
flexiverse
|
|
« Reply #89 on: July 09, 2015, 04:53:21 AM » |
|
I did it! !!!!! whuaaaaahhhhhh! I got it set up! my workstation is dreaming right now!!!!! phew. took my the whole afternoon... It's coommandline stuff and I have no idea what I am doing and where all that 4gb of stuff installed itself to, but its working! not comfortable at all. but does the job! I don't know the last steps exactly and what lead to it finally working, but if you try this guys version, it obviosly is manageable for a non-coder like me: https://github.com/Dhar/image-dreamertry at your own risk!! and feel free to ask me, maybe I recall what I did when I was stuck at the same point - probably not.. ---- wow. this is slow! and it doesn't solve my problem of how to do larger resolutions than 1024*1024. Seems like memory kills it. damn. doesn't make use of my 16gb ram. why not?! aww.. all this for nothing?! well, at least I can continue to dream low res without the website.. Just use fractal compression to enlarge to any size withiout loss. With these images fractal compression is perfect to create higher resolution versions. |
|
|
|
|
Logged
|
|
|
|
|
September 06, 2011, 04:07:55 PM
Dilber MC Theme by HarzeM