Ai Dreams Forum
Artificial Intelligence => General AI Discussion => Topic started by: LOCKSUIT on April 07, 2018, 01:56:47 am
-
If Ian Goodfellow invented GANs in 2014 then what is this 2011 DARPA video?
Ian Goodfellow invented GANs in 2014:
https://en.wikipedia.org/wiki/Generative_adversarial_network#cite_ref-DAGUI_4-0
2011 DARPA video showing a excellent GAN-like use:
https://commons.wikimedia.org/wiki/File:DARPA_Visual_Media_Reasoning_Concept_Video.ogv
FINALLY A GOOD SUPER-RESOLUTION APP! CAUTION EACH EMAIL HAS ONLY 5 TRYS! MUST PAY! https://letsenhance.io/boost
oh i know ._. just make a collage of images together and feed it in as if it's 1 image OH SO EVIL MEHAWAHA now u guys O me !!!!!!!!!!!!!!!!!!!!!!! THINK OF IT AS A AI DREAMS FORUM PROMO CODE
Scroll to bottom to see horse to zenbra video...wow this redefines "zebra" (inside joke).
http://geoffreylitt.com/2017/06/04/enhance-upscaling-images-with-generative-adversarial-neural-networks.html
In other news whaaa noohh DARPA has used this already for crime fighting, it certainty helps!!
"The low-resolution images obviously don’t contain enough information to actually reconstruct the original image. So to some extent, our neural network is just making up new pixels that seem realistic. This technique may be ready for TV, but it’s not ready for actual crimefighting."
-
AND OMG IT WORKS OH WOW I FEED IN A 2000X2000 COLLAGE ANNDDDD IT FEEEDS ME OUT A 8000X8000
OHHHHHHHH I AMM SOOOOO HAPPYYYYYYYYYYYYYYYYYYYYY !!!!!!!!!!!!
-
HERE IS THE INPUT
https://advancessss.deviantart.com/art/Collage-1-739212705
-
HERE IS THE OUTPUT
https://advancessss.deviantart.com/art/Collage-1-magic-739212571?ga_submit_new=10%3A1523066533
AMAZING! I WONDER HOW MANY UPLARGES I CAN DO! THE FUTURE IS SO FUN!!!!!
-
DOWNLOAD BOTH IMAGES AND ZOOM IN UNTIL BOTH AT SAME SIZE TO SEE THE DIFFERENCE LIKE THIS:
https://advancessss.deviantart.com/art/WOWsuper-resolution-739213959?ga_submit_new=10%3A1523067486
-
note you can feed in a 5 megapixel image limit
-
Here is a hack way to impleent it yourself
https://hackernoon.com/how-to-replicate-lets-enhance-service-without-even-coding-3b6b31b0fa2e
anyone able to give us a free version? We deserve it :P
-
btw guys i found more freeeeeeeeeee ones that are good wait a moment
also https://www.digitaltrends.com/cool-tech/algorithm-low-res-high-res/
here is it, the best one currently possibly: http://webdav.tuebingen.mpg.de/pixel/enhancenet/
It's easy to setup it's pre-trained n all. Can anyone film a vid?
-
man been at this all day. 9 hours on a chair. here is several pic AND vid !!! holeemoli folks!!!!!! Best FREE program is is at top.
VIDEO
(fix url) https ://youtu.be/QKzqlZ_cKEQ --------- COLLAGE OF THE BEST, READ DESCRIPTION FOR PROGRAMS!!!
http://www.infognition.com/videoenhancer/ ---------- YER
PICTURE
https://www.isize.co/b2c/ & https://www.isize.co --------------- FREE U CAN DO LOTS.....LOOK AT THE FRONT PAGE!!!
https://bigjpg.com/ ---------------- U R GUST GORGISH BABY
https://letsenhance.io/boost ------------ I LIKE U TOO BUT YOU WANT MONEY SO...
https://waifu2x.booru.pics/ --------------- U R MY WAIFU HONEY BABY AND YOU WORK GREAT
http://reshade.com/
http://a-sharper-scaling.com/
http://www.imageenlarger.com/
-
Well I'm done with this, for now.
I found Let's Enhance is best but the free ones iSize AnimerOnelol and the AI upscaler to be very same and good essentially.
I believe infinite super-resolution is possible. A monkey-in-a-black-room could do it - I could draw up one if given a few hundred years 20x its size. We KNOW if it looks right. Just think, we have it in us trained and know. That's proof infinite SR works.
'Tada'
.....ok guys seriosly now about the thread question....what gives?
-
If Ian Goodfellow invented GANs in 2014 then what is this 2011 DARPA video?
Ian Goodfellow invented GANs in 2014:
https://en.wikipedia.org/wiki/Generative_adversarial_network#cite_ref-DAGUI_4-0
2011 DARPA video showing a excellent GAN-like use:
https://commons.wikimedia.org/wiki/File:DARPA_Visual_Media_Reasoning_Concept_Video.ogv
FINALLY A GOOD SUPER-RESOLUTION APP! CAUTION EACH EMAIL HAS ONLY 5 TRYS! MUST PAY! https://letsenhance.io/boost
oh i know ._. just make a collage of images together and feed it in as if it's 1 image OH SO EVIL MEHAWAHA now u guys O me !!!!!!!!!!!!!!!!!!!!!!! THINK OF IT AS A AI DREAMS FORUM PROMO CODE
Scroll to bottom to see horse to zenbra video...wow this redefines "zebra" (inside joke).
http://geoffreylitt.com/2017/06/04/enhance-upscaling-images-with-generative-adversarial-neural-networks.html
In other news whaaa noohh DARPA has used this already for crime fighting, it certainty helps!!
"The low-resolution images obviously don’t contain enough information to actually reconstruct the original image. So to some extent, our neural network is just making up new pixels that seem realistic. This technique may be ready for TV, but it’s not ready for actual crimefighting."
The DARPA video is a Visual Reasoning Media concept video. It did not mention it being GAN-like. You did. It and other concept experiments were done in the early 1990's and on until someone decided to perhaps give it a moniker GAN. The results are pretty amazing as to what information and how much info can be extracted from a single video!
Nice find Lock! O0
-
It was pretty tricky to do the following because the enlargers above can't do that much idkwtf not but anyhow I surpassed their limitations and made a fairly large super-resolution better than the freaking original image of a bird look:
(also attached is the full size image showing off how good i got it. i had to use different editing tools AND different superresolution-ers above:
-
Here's a cool underwater one. It likes bigger images btw.
ORIGINAL
https://advancessss.deviantart.com/art/151230224432-aequorea-49-super-169-739740485?ga_submit_new=10%3A1523356892
SUPER FAR ZOOM SR
https://advancessss.deviantart.com/art/37773959cbc70878ede265593d77700424968c21-S2-N1-2x-739740754?ga_submit_new=10%3A1523357251
FURTHEST ZOOM SR - VERY SIMILAR TO THE ABOVE - IT'S BETTER.. - BUT CROPPED - SORRY BOUT THAT
https://advancessss.deviantart.com/art/5-E0a23747076d92f0983803cb34da50a1db2707f9-S2-N1-739740520?ga_submit_new=10%3A1523356929
-
This one is cool. It's the Tesla car factory upsized a lot.
original
https://advancessss.deviantart.com/art/Tesla-fremont-factory-Large-739892108?ga_submit_new=10%3A1523448632
super-resolution
https://advancessss.deviantart.com/art/fab-lab-2x-2iNoise-2x-2iNoise-739892173?ga_submit_new=10%3A1523449069
-
It’s cool but it’s just a modern algorithm/ take on interpolation.
You can’t get something from nothing.
New detail cannot be added, if it wasn't in the original image.
:)
-
It’s cool but it’s just a modern algorithm/ take on interpolation.
You can’t get something from nothing.
New detail cannot be added, if it wasn't in the original image.
:)
Unless you use neural networks backwards to generate an enlarged image.
-
Oh korrelan you are so wrong.
If a bike in an image says a certain make/brand on it, an ASI can refine the image using super duper methods and make everything ultra clear. It can look at the humidity levels, etc. Just like they did in that DARPA video, but a million times more advanced. JUST LOOK AT HOW MUCH INFO THEY PULLED IN THAT DARPA VID
ivan....um....we get that...lol...........I used the GAN on the "Let's Enhance" website.
hey guys why is the darpa robots surprising they are just scripted to do backflips....we already have robots, now we just need AI. Sure they can track etc NARROW but backflip??? oh come on, why...? Their robots are'nt impresive because they'll have pretty narrow use until the holy spirits fills them.