Imagine if in the future, prototyping levels for video games was as easy as a sloppy 5 minute sketch
Imagine if this AI was fed all the data from Google Street View
Holy shit that is actually fucking cool as hell.
That wouldn't work properly because its far too complex. This is a simple GQN that can barely handle simple geometry like this at the moment. Its state of the art, but that doesn't mean we've reached anywhere near the level of understanding something like Street View.
Let's be honest - it probably is already. :P
Pair that with VR and you can make traveling the world SO much cheaper
It'll be cool to see this being advanced enough to construct future crime scenes or past crime scenes.
This is insane. The possibilities for this are endless. Neural network science is moving mountains right now.
Seems cool but I think it is a little bit sensationalist. It doesn't "imagine worlds".
I creates 3D spaces from detecting shapes in the given image. 2nd Image in OP actually shows what it does.
Its still pretty impressive and I wonder if this could potentially help with things that are currently done with LIDAR for example.
Of course not, but it gives me hope of something of that sort being within the scope of possibility in the near future. Hell, even the Street View as we have it now is a pretty crazy feat that I never would've imagine existing 20 years ago. Deepmind gives me a lot of hope for some crazy shit, and Google have the power to make that crazy shit a reality.
I'm sitting here talking about this to my friend. Does this mean computers can dream now? Sometimes when I sleep I dream up buildings or open spaces that are unique but based on pictures I've seen or places I've been.
Obviously this is at a very crude stage, but couldn't this technique eventually be applied to allow ai to dream up unique art inspired from a real life photo?
If by "dreaming" you mean creating totally new information based off of pieces of real data, then sure.
But beyond a semantic debate about what your definition of dreaming is, no I think you're giving this a bit more credit than its due.
Not using the current technology. What they do is generate thousands of computer generated scenarios to train the neural network with, and then apply it to an image generated the same way as the training data. So all it will really do is generate data that is similar to what it has seen before.
It's pretty much like a more sophisticated version of those markov chain text generators where you input some text and it generates more text that has the same style.
We will know when we give it a street view imagine and it comes back rendered with hundreds of screaming human corpses.
Why does the Neural rendering look like it was rendered with ncurses
20-30 years.
All i have to do to make a game is just draw.
I want to see what this could do with Wolfenstein 3D screenshots.
will this make anime real?
can you deepfake with this?
no
?
No?
No, that's a different thing altogether. GQN's wouldn't be used for that at all, I wouldn't think.
https://d1o50x50snmhul.cloudfront.net/wp-content/uploads/2018/06/14142649/deepmind-maze-gif.gif
Super impressive. I hope we'll have access to trying it for ourselves sometime.
HDTF 2: Electric Boogaloo coming soon.
Does this mean someday AI can be creatives? "Write a symphony, turn a canvas into a beautiful masterpiece" and all that?
I wonder what it would do if you provided it with stills from porn
Well I mean they already can, but the quality is a matter of taste.
https://puu.sh/AGlOY/2a30b2e146.png
It didn't recognize a child...
... skynet lives.
Depends what you consider "creativity". Machines have long been able to generate music and pictures from scratch. Does that count?
Genetic algorithms can produce solutions to complex, abstract problems. That's pretty creative when you think about it
Sorry, you need to Log In to post a reply to this thread.
