Stanford's New Website Converts Your Photos to 3D

An anonymous reader writes to tell us that Stanford has a new website that not only shows you how cool their new 3-d modeling system is, but actually allows you to give it a try with your own photos. The system can take a 2-d still image and estimate a detailed 3-d structure which you can navigate. "For each small homogeneous patch in the image, we use a Markov Random Field (MRF) to infer a set of "plane parameters" that capture both the 3-d location and 3-d orientation of the patch. The MRF, trained via supervised learning, models both image depth cues as well as the relationships between different parts of the image. Other than assuming that the environment is made up of a number of small planes, our model makes no explicit assumptions about the structure of the scene; this enables the algorithm to capture much more detailed 3-d structure than does prior art (such as Saxena et al., 2005, Delage et al., 2005, and Hoiem et el., 2005), and also give a much richer experience in the 3-d flythroughs created using image-based rendering, even for scenes with significant non-vertical structure."

Comments

Slashdotted

[d3ac0n]

Aaaaaand it's already slashdotted.

Wow. That was fast.

Re:Slashdotted

[alta]

No kidding, who would have ever thought that putting a link to /. to a service that does IMAGE PROCESSING was a good idea. Image processing is intensive on any server. Hell, lately /. can't even handle /.'s loads. It took 2 minutes to load this comment page, and earlier I was getting 300 errors when trying to read comments!

Re:Slashdotted

[Anonymous Coward]

http://www.youtube.com/watch?v=GWWIn29ZV4Q

Yes, the website was slash-dotted.....
We plan to have it up again at: http://make3d.stanford.edu/
(Slashdot brought the whole Stanford AI lab servers down: http://ai.stanford.edu/

Meanwhile, please see:
http://cs.stanford.edu/people/asaxena/reconstruction3d/
for an year old page.

Re:Used for navigation systems?

[grub]

Could this type of technology be used for robots to allow them to identify what the 3d layout of the world around them is?

Some (most?) robots already use dual cameras for true depth perception.

Re:Used for navigation systems?

[moderatorrater]

The problem is that binocular vision get's less accurate at longer distances. Also, for whatever reason, the robot might not be able to use two "eyes". Either way, another method of approximating distance would come in useful for anything that gets a lot of every day use.

Photosynth

[blankinthefill]

While I know you're all Microsoft haters, bear with me for a minute. This sounds a lot like this Photosynth demonstration. The relevant part of the video starts at about 3:50, but the whole video is really interesting and I would suggest watching it.

Re:Photosynth

[nguy]

Photosynth takes multiple shots, this apparently takes a single shot. And although Photosynth is some nice engineering, (1) it wasn't all developed at Microsoft, and (2) it relies on decades of research work done elsewhere.

Microsoft does invest a lot of money in research. But what they are spending pales in comparison to all the work by other people that they are building on.

Re:Photosynth

[pcgabe]

Photosynth doesn't make anything 3D. It combines flat photos, and while you can move around and see photos attached at different angles, each of those views MUST be a photo on its own. The more pictures you add, the more angles you can look at, but Photosynth isn't making anything 3D.

These two packages are quite, QUITE different.

Re:Used for navigation systems?

[disckitty]

Judging from the google cached pages, it looks like that's precisely what his research is for. Google cached pages: here, and here, and here

It's not slashdotted.

[Half-pint HAL]

I'm sorry to say that us geeks have been usurped by young hipsters in the website-disabling stakes. This site has not been slashdotted, it has been YouTubed. Someone at Stanford has been uploading videos of this to YouTube and inviting the plebs to go to their site before us. How ungrateful. The swines. Harumph.

HAL.

Re:It's not slashdotted.

[longacre]

I'm a \. subscriber and saw this story about 2 minutes before it went live to the masses. In that time, I was able to successfully visit the site and register. By the time I logged in, however, the \. post had gone live, and the Stanford site stopped working altogether. So it was indeed \. that crashed the site, not YouTube.

Re:Games?

[mwigmani]

Student Arrested for Making Videogame Map of School