Now available in 3D!
September 9, 2013 11:25 AM   Subscribe

"We introduce an interactive technique for modelling 3D man-made objects by extracting them from a single photograph." 3-Sweep: Extracting Editable Objects from a Single Photo, SIGGRAPH Asia 2013 (SLYT)
posted by oulipian (17 comments total) 25 users marked this as a favorite
SIGGRAPH demos never fail to completely blow my mind.
posted by jason_steakums at 11:34 AM on September 9, 2013 [3 favorites]

Whenever I see these SIGGRAPH demos, I feel like I'm living in the future. It's almost too much.
posted by Admiral Haddock at 11:37 AM on September 9, 2013 [2 favorites]

posted by echo target at 11:43 AM on September 9, 2013 [1 favorite]

While I'm always blown away by things like this, I sometimes have an equally opposite reaction: in this case, I feel like I'm caught halfway between "OMG where will reality end and CGI begin" and watching part of a documentary of the making of the hypothetical Lawnmower Man 2.
posted by not_on_display at 11:48 AM on September 9, 2013 [1 favorite]

Looks like Cyriak's life is a lot easier now...
posted by pipeski at 11:53 AM on September 9, 2013 [7 favorites]

The basic techniques (3D shape from image extraction) that seem to be shown by this video have been known for years, although the UI and shown speed are quite nice. Unfortunately, since this is currently an ACM SIGGRAPH submission, we can't see the actual paper to learn anything about what is potentially new here.
posted by riotnrrd at 11:56 AM on September 9, 2013

This would be such a boon to indie game design. Churning out all the incidental objects in a scene is such a time suck for a single developer.
posted by jason_steakums at 12:11 PM on September 9, 2013

Texture mapping with no lighting makes anything look good... I want to see the meshes up close.
posted by scose at 12:32 PM on September 9, 2013

Brilliant! I remember messing around with Canoma when I was younger, thinking it was going to be a huge thing. Then Adobe killed it.

I'd be really interested to see this used with game concept art. Forget "it looks like the concept art", it is the concept art.
posted by lucidium at 12:34 PM on September 9, 2013 [1 favorite]

posted by aaronetc at 1:25 PM on September 9, 2013 [2 favorites]

Note that you can't change the viewpoint or light direction freely in these models. Using symmetry to fill in the unseen part of the object is a neat trick but obviously does not work for most objects.

If you have the object at hand and not just a single image, Autodesk 123D Catch might interest you.

In practice, I'd say that the problem of getting concept art or incidental objects in a scene is mostly solved by Unity asset store. I mean this seriously. Large collections of 3D models are dirt cheap there. If you're an indie game developer and not a proficient 3D modeler, don't waste any of your time doing it yourself.
posted by ikalliom at 1:26 PM on September 9, 2013 [1 favorite]

Enhance 57 to 19. Track 45 left. Stop. Enhance 15 to 23. Give me a hard copy right there.
posted by ricochet biscuit at 1:36 PM on September 9, 2013 [4 favorites]

ikalliom: "Using symmetry to fill in the unseen part of the object is a neat trick but obviously does not work for most objects."

Yeah, it seems like it places very tight restrictions on what kind of shapes you can work with in that it requires you to find a circular or rectangular/square planar shape to extrude from (triangles won't work because one edge could be fully obscured and thus not inferred from what you can see). And it requires objects to have rotational symmetry around the curve along which you extrude (which is why it fails on the toothpaste tube). Is it a quick, intuitive interface and a cool little demo? Yes. But, given its limitations, what could it be actually useful for?

Autodesk's 123D Catch is way more flexible and impressive in my opinion. And you can get productivity out of it as well. I used it for quickly making 3d models of oddly shaped things for an interior design project. It would have taken me forever to make those models by hand based on measurements and it would have been completely impossible using 3-Sweep.
posted by Hairy Lobster at 1:57 PM on September 9, 2013 [1 favorite]

Enhance! Enhance!

Rotate 60 degrees!
posted by Mercaptan at 2:05 PM on September 9, 2013 [2 favorites]

I would love to get something like this that integrates with the compositing programs I use - there are so many times that I have to fix or tweak what has been filmed - like, someone throws a knife, and they want it to be aimed better. It would be great if I could grab the image of the knife right from the scene, and have some dimentionallity to it as I move it around. Preserving the lighting and image quality it already has would actually be a big plus.
posted by 5_13_23_42_69_666 at 2:59 PM on September 9, 2013

Ummmmmm, yar.

Not sure much else to say but I hope they make this readily available at a consumerish level.

(Yar my word to say when I can't think of a word to say. I supposed you might say I am gobsmacked right now.)
posted by Samizdata at 4:25 PM on September 9, 2013

This is some truly amazing tech... but at the same time, it seems like every few years a video turns up online demonstrating some revolutionary new program that can extract 3D objects from photos. It's always dazzling, and seems like something that could have tremendous applications in all sorts of fields... and then it goes away and I never hear about it again. Here's hoping this isn't another one of those.
posted by Ursula Hitler at 5:48 PM on September 9, 2013

« Older the courage to laugh back   |   Using cash is expensive - especially if you're... Newer »

This thread has been archived and is closed to new comments