DeepMind AI can "imagine" a world based on a single picture - Sensationalist Headlines

while I was joking, the only idea coming to mind is reconstructing parts of a face without glasses. I have no interest in using it with people, but I'm curious on how well it would handle high dimension sets of data. Also here's some videos linked from the paper https://www.youtube.com/watch?v=G-kWNQJ4idw https://www.youtube.com/watch?v=RBJFngN33Qo

Give it screenshots of levels from games that got never released but managed to get some pics in some sort of media like trailer or game news site or something Let it recreate what never existed

The biggest thing I take away from this is translating real world things into level design, allowing you to quickly create a simple level with boxes and legos etc and then recreating it into a game. You could recreate a level using a very simple and small scene you built using 3D printed objects that you then paint. Imagine combining this with voxels, meaning your levels will be totally unique every time. Take a bunch of photo's of parks and then use those in the video game, bam.

This isn't nearly complex enough for that; also, what's wrong with our current methods?

3D printed might be a bit silly, though, the idea is that you have a 'real version' of everything that you can look at irl and physically get an idea of. I love seeing video game levels being recreated in real life. It'd be cool to create a 3D model of a bunch of random objects, and then turn it into a scene, and recreate it digitally to see how accurate it gets it to the 'original'. Better than current methods because they don't build the rest themselves, and instead rebuild only what they see. I see tons of photo-generated levels in 'Vive home', they're cool but they require a ton of work and you end up with a huge mess of errors and really awful looking objects.

Well with 3D scanning, sure.. but games don't really use that to make environments. They many scan individual objects to place on a handmade environment (although usually not) but I've never heard of entire environments being 3d scanned in. High quality work in this area is still squarely in the hands of artists.

this could hypothetically lead to a point where you can have an HD 3D model automatically rendered for you just by taking a few pictures of a real life object

Really wanna see a publically useable demo of this just to see what kind of things we can do with it.

Uh... https://youtu.be/8M_-lSYqACo

Only issue with 3D scanning is that unless you have an insanely good camera it's really crappy looking and creates a huge number of flaws. it also needs a lot of pictures at a good angle. that isn't to say this will be better and all that, but it could be used in combination with 3D scanning to 'recreate' the parts it has to create

It's so funny because i used to be all judgmental and like "this is not how this shit works! The fuck is up with the writing!" Well joke's on me! This IS how it works!