Extracting 3D objects from a photo and then editing them right away

[leonid]

This stuff is black magic

[noname219]

At first I was laughing.

But then, I got scared. This is crazy.

[kommisar]

and soon will be printed with a 3d printer

[andy-o24]

Photography can no longer be trusted. I'll watch it with volume later.

-o24

[Hateandhatred]

Holy fucking shit

Now I want to take a picture of a naked human from the front and rotate it and see what this soft will put in the human's back

[Crazyjayde]

This might get actually scary with human models.

[Dynam0]

It obviously can't 'predict' what's on the other side of things. If you were to do this with a six-sided die showing the '1', '4', and '5' face for instance, it would assign a dimension to each face and make a copy of it on the other side. Rotating the image of your cube with this software would give you a die showing 1, 1, 4, 4, 5, and 5. You obviously can't make a 2-dimensional image 3d without some flaws

Notice how with the toothpaste tube it assumes the scaling is uniform; since the width doesn't change, it assumes the depth doesn't change as well so the tail of the toothpaste is lost. For conical objects, the width and depth would change at the same rate so no detail would be lost there.

Still insanely useful software..wowza

[Hateandhatred]

Originally Posted by Dynam0

It obviously can't 'predict' what's on the other side of things. If you were to do this with a six-sided die showing the '1', '4', and '5' face for instance, it would assign a dimension to each face and make a copy of it on the other side. Rotating the image of your cube with this software would give you a die showing 1, 1, 4, 4, 5, and 5. You obviously can't make a 2-dimensional image 3d without some flaws

Notice how with the toothpaste tube it assumes the scaling is uniform; since the width doesn't change, it assumes the depth doesn't change as well so the tail of the toothpaste is lost. For conical objects, the width and depth would change at the same rate so no detail would be lost there.

Still insanely useful software..wowza

That's why it would be so amazing to do it with humans rofl

[SKG_Scintill]

so the second to last picture, how does it figure out what the tree looks like when the original photo has a piece in front of it?

[Crazyjayde]

It's a wonderful tool for concept art artists though, as the telescope portion of the video shows. Of course, it's still not the bridge between 2D to 3D modelling (meaning, it won't give you the entire model just with one click), but it has some nice applications.

@SKG: It tries to fill the background approximately by copying adjacent elements. Somebody could probaby refresh my memory because I can't remember the name of the process, but 4:30 could give you a better example of what it can do wrong. Also notice where the chandelier stands at 3:45.

[ReikonKeiri]

Fuuuuuuuuture

[Arkuski]

Originally Posted by Hateandhatred

Holy fucking shit

Now I want to take a picture of a naked human from the front and rotate it and see what this soft will put in the human's back

We all know this will be on the other side of the face:

[noname219]

Originally Posted by Crazyjayde

@SKG: It tries to fill the background approximately by copying adjacent elements. Somebody could probaby refresh my memory because I can't remember the name of the process

I think it's called Content-Aware Fill or at least, it uses a similar method.

[Hateandhatred]

Originally Posted by Arkuski

We all know this will be on the other side of the face:

Yeah I really hope this will happen.

If so, time to mess with friend pictures on Facebook.

[MixMasterLar]

Annnnnnnd Photos now mean shit.



Welcome to the future

[mi40]

if u take a photo of a human's ass and rotate it with this software do you get this

[Pseudo Enigma]

What is this sorcery

[Nullifidian]

This looks cool at first glance but I see a limited use for this tool other than improvement in workflow for quick concepting of geometric shapes or simple photo editing, and even then it's very limited in what you can and can't change. Anything unique, intricate or organically shaped would likely be a nope (as you could tell from the error result of copying a simple tube of toothpaste).
These models won't be of much use for actual assets to be used in a game, movie or whatever either since no computer generation will make optimized models or texturemaps (be it for further editing or final assets) and wouldn't be able to take into account various properties the texture should have (think gloss, reflection, microdetail such as cracks or bumps etc.). These would still require a human hand to be recognized. As one guy in the video comments already said this will improve workflow, but is not going to revolutionize the industry.

edit:
If you want to look at something revolutionary, you should take a look at some of the demos that are being put out for the oculus rift. The possibilities with that thing...
Take a look at this demo for example:

[mi40]

wait wtf that wasn't a joke video?

[Nullifidian]

The oculus rift one? It sure wasn't