Any tools that can estimate 3d shape based on lighting of 2d image?

[WesIsGood]

Just wondering if this existed. For all I know, there's an easy way to accomplish the same thing in photoshop or whatever, but I was just curious if there were any tools that would analyze an image, look for lighting patterns, and try to guess the 3d shape of the things in the image. And then, allow you to change the light source, and attempt to recreate the image with the new light source in mind. For simple things. Obviously mirrors and semi-opaque glass and whatnot would mess everything up.

[hatu]

I saw a video about that not too long ago, I dunno if they've released the program yet

http://hothardware.com/News/Capturin...-Flash-Camera/

[renderhjs]

Autodesk ImageModeler is a application designed for IBMR (image-based modeling and rendering)
http://en.wikipedia.org/wiki/Image-b..._and_rendering
the autodesk product page is:
http://imagemodeler.realviz.com/phot...-software.html

I tried it cone myself but honestly I think the navigation and interface are somewhat bad designed - there is alot of linear workflow (something unknown in Flash or 3dsmax)- and it therefore often gives the feeling of a slow pace.



Of course you can do image Modeling in Maya, 3dsmax, XSI or the free blender as well (I´ve seen examples in all of them before). For example Gnomon lists some video lectures on their site:
http://www.thegnomonworkshop.com/dvds/gdo02.html
http://www.thegnomonworkshop.com/dvds/gdo03.html
Btw. many of the Studio Ghibli (Japanese anime- studio) used Softimage XSI (a 3d application) to map 2 painted backgrounds on 3d models and with that fake 3d even though it looked completely 2d in its style.
In the movie Industry its standard and used alot for the digital matte- paintings - for example to extend a western town that isn´t even build on the set - or take King Kong where they completely remodeled the NY city in 1930 - lots of techniques involved camera mapping.
Not so long ago I did a camera mapping experiment in 3dsmax with a photo from london at night, you can see some picture of how it was done here:
http://boards.polycount.net/showpost...postcount=1240

Then you have video trace
http://www.acvt.com.au/research/videotrace/
Which is like image based modeling but applied to video frames. This gives you some advantages like the ability to adjust vertex points on frames (while the system tries it best to do it automaticly).
In the end you can texture- map a object easier as for example ImageModeler,- but right now there hasn´t been released any software pieces from them.

In a own research I started not so long ago on a texture ripper that can extract rectanglur textures from distorted perspective images or photos
http://boards.polycount.net/showthread.php?t=54694
that thread lits a bunch of other commercial tools that do the same but way different as my approach.

but pure based on lighting inside a single photo is not something I noticed so far- most attempts are technologies that want to put you in some control to tweak things and give it a persoanal or artistic direction (that´s what sells for the professional industry, not automatic stuff with crappy results).

[VENGEANCE MX]

Microsoft Photosynth does a pretty good job of creating a composite 3D navigable environment from a bunch of photos. For one demonstration, they took every photo on Flickr tagged 'Notre Dame' (different sizes, angles resolutions, times of day, camera, exposure, blur, zoom; you get the picture... if you'll pardon the pun), dropped them all into the algorithm and it created a hi-res fully explorable rendering of the Notre Dame building (obviously only the outside, though).