3-Sweep: Extracting Editable Objects from a Single Photo

Brygelsmack

What do we think of this?

Pretty cool tech but I don't see it replacing any of the tools we use today. Not now anyways.

[ame="http://www.youtube.com/watch?v=Oie1ZXWceqM"]3-Sweep: Extracting Editable Objects from a Single Photo, SIGGRAPH ASIA 2013 - YouTube[/ame]

JamesArk

Yeah, it's a very cool start but is still clearly limited right now, even by examples in the video. Everything is a lathed or perfectly symmetrical object. As a plug-in for a Max/Maya type of package it would be very cool and in a few iterations would be great for blocking stuff out if it could handle more complex shapes.

Obviously topology and the texture quality are big issues but it's still quite impressive for what it's aimed at.

ExcessiveZero

Nifty little tool looks like, not exactly ground breaking and its nice they point out their issues, but could be great for base meshes and basic geo.

WarrenM

Not exactly ground breaking? I think there's a number of things here that are very notable. Most particularly, the automatic following of the shapes and using perspective guides to produce the correct volume in the shape.

I can see this leading to some very cool places - Not the least of which making concept art REALLY easy to translate into base geometry for the modelers to start from.

Fuzzy_Bison

Complete rookie here, but I'm trying to find more info on this 3-sweep. I've gotta ask, Brygelsmack says

Brygelsmack wrote: »

...

Pretty cool tech but I don't see it replacing any of the tools we use today. Not now anyways.
/QUOTE]

so I gotta ask, What tools are you using that render 3d from a photo? Anything as simple as this (appears)?

I see that there are limitations, but, for sure, this appears simple enough for the average user to handle. Is something this simple already out there?

Also, some replies on the video are saying it's fake. Thoughts?

Noors

Just checked, Teddy is 14 years old
[ame="http://www.youtube.com/watch?v=e2H35SlLmUA"]T.E.D.D.Y. (Draw in 2D - Outputs in 3D) - YouTube[/ame]

So not exactly ground breaking ^^ Yeah, looks like it only extrudes a square or circle so the perspective can't really go wrong. Could have some utility tho.

Vailias

Not new,no. This tech goes back to 1994 at least. This one is neat though, and its texture approximation for the backfaces is pretty good.
The biggest integrated feature I can see in that demo video is the automatic image background patching ala Photoshop's Content Aware Fill.

Fuzzy_Bison: Tools for this sorta thing:
http://en.wikipedia.org/wiki/Photogrammetry

AlexCatMasterSupreme

I want to model a potato.

Fuzzy_Bison

Thanks for the replies.

I'm looking through the Wikipedia list (I've looked at 7 or 8 so far) and I still haven't found anything that is quite like the demo in the first video. A 3d editable object created from a single photo. The ones I've looked at all require multiple overlapping photos.

Like I say, I'm a rookie here and at the whole concept of 3d modeling so it was the simplicity of 3d from 2d that caught my eye.

Teddy may have been a precursor. It's still up on the web at http://www-ui.is.s.u-tokyo.ac.jp/~takeo/teddy/teddy/teddy.html but I can't find any direct evolution from this ancient (1999) ancestor.

Thanks again to all.

TheWinterLord

Its pretty cool, its almost more cool however, that it knows how the clouds look like behind the binoculars after tweaking the model around...

Vailias

Notice how they didn't actually make anything complex in the demos.
They exclusively stuck to things that are loft objects, as perna says, essentially a bunch of cylinders.
Try a car, or anything with even mildly complex and it will break down hard. They even show as much with the tube of toothpaste.

Also you won't find anything exactly like the video for the subject software. If it was Exactly doable with existing tools they wouldn't have written this new one.

Its neat, its novel in that it uses intelligent image patching and texture generation for occluded surfaces of the object rather than straight projection mapping, but it's essentially useless for anything in a professional capacity.

ENODMI

AlexCatMasterSupreme wrote: »

I want to model a potato.

Too bad!

Accipiter

Vailias wrote: »

They even show as much with the tube of toothpaste.

If you listened to it though, they were talking about their complications with the tube of toothpaste not being able to create that flat area in one stroke. Also the glass bottle that (I'm guessing) was underwater because of the fuzzy edges of the bottle itself.

If all you need is around 3 strokes for something, then do an export, I'd say that could be pretty useful. However I suppose it could come down to topology. But maybe with a little Zremesher help? lol

McGreed

Yeah, I believe it's only useful for blocking out stuff, and that's about it. Also, how does it account for FOV/Lens in the picture?

David-J

I agree it seems rather limited when it comes to complex shaped but for matte painting I think it cold prove useful.

Vailias

Acciptier: Yeah I listened. The point is you're adding an extra step and piece of software to a blockout process, giving up control, and introducing error.
What is the upside? That you can generate crap to clean up faster? That you can make tube like objects quickly?
You're not going to keep their projection mapped, patch filled textures, which is a good half of what makes the models look at all presentable.

So while this is cute, its still generally useless to this particular industry. It will likely have a nice niche market in image manipulation and augmentation.

Noors

Yeah could be inside photoshop, but in a 3d pipeline, i don't think so.

ZacD

I could see it being nice if you have to make dozens of pots or bottles from reference. but otherwise its very niche use.

Accipiter

Vailias wrote: »

That you can make tube like objects quickly?

Really made me laugh out loud.

Yeah, I suppose not in a 3D production pipeline, but perhaps for some concept type stuff in Photoshop.

Justin Meisse

perna wrote: »

First the user manually defines 3D space (it's NOT automatic), then the rest is more or less just a simple extrude based on detected edges.

ugh, that's the biggest let down. A program that automagically attempts to figure out the perspective of a scene and possibly compensates for lens distortion would be awesome. So far, Sketchup has come the closest but you still have to fiddle with perspective lines first.

MainManiac

Would have been great for a game with the same quality as GTA 3/Vice city, I can see it being used for tons of detail objects

Blaizer

this may be very useful using hi-res photos, it's just another photo to 3d app. For game art... it may be very counterproductive. BUT, at least, we can block out a model with real proportions in seconds and obtain textures without any effort. The bigger the photo source, the better.

http://insight3d.sourceforge.net/
http://www.3dsom.com/
http://www.strata.com/products/strata_3d_cx_suite/strata_foto_3d_cx/
etc.

Search in google: "3d from photos"

Vailias

Blaizer: Yeah photogrammetry is a great technology with a lot of uses.
There is nothing wrong with using a point cloud generator to get a high res mesh from photographic data, especially if the algorithms are of good quality, ala photomodeler, or most of the 3d optical scanning methods. That DOES have an application to games and is used. see: http://www.gamasutra.com/view/news/103960/Eyetronics_3D_Scanning_Used_For_Mass_Effect.php for one example

Its more that this specific program seems limited to lofting and texture projection. (i.e. stuff you could do in minutes in mid 90's ddc apps with a little setup of a reference plane to work with.)