By Cory Doctorow at 2:16 pm Thu, Dec 2, 2010
A UNC team has written an engine that scours Flickr for photos of a city, figures out which ones are images of the same place, analyzes them, and uses the results to build amazingly detailed 3D models -- all in less than a day, using a single PC.
Flickr Hack Makes 3D Model of Any City in a Day
(Thanks, MooseHP, via Submitterator)
Wow that is impressive. The future just got closer.
Imagine the next step: doing this with surveillance cameras. Live.
How are the final models stored? They look like continuous surfaces, but usually these types of picture->3D algorithms use voxels, which results in a more blocky look.
The details caught in statues are especially interesting.
It would be interesting to try to pull similar data out of the Google Streetview data.
Obligatory “Rome wasn’t built in a day” joke.
Here’s the paper: http://cs.unc.edu/~jmf/publications/Frahm_et_al_ReconstructionFromPhotoCollection.pdf
They’re reconstructing a polygonal surface representation (i.e. lots of triangles).
describing something as taking “less than a day, using a single [computer]” puts me very much in mind of computing in the 1960s..
And means that your kids will be doing it in realtime and not find it the least impressive.
You know when you go to a concert and there are a million people holding up their cellphones and filming the show from a million angles?
It’s gonna be awesome when some future music historian downloads them all from youtube, syncs the audio tracks and crunches them into a 3D virtual environment. Walk around, see people in the crowd, relive the event as if you were really there!
iPhone/smart camera application suggestion:
Suggest locations to photographers at famous locations where they should shoot from to contribute the most new information to the 3d models of these locations. Nearly all of the shots are from the same locations. Some variety would help enormously.
Is it much different than Photosynth?
How is this different from Photosynth?
How is this different from Photosynth? http://www.youtube.com/watch?v=M-8k8GEGZPM
It’ll be something once they unleash it on porn.
Someone should run this on Google Earth Engine, and create 3D models of the entire world.
Really cool! This screams for crowd-activating follow up projects. Declare some week “3D our town week”. On that day, people are encouraged to go about town and take photos and then upload them to flickr or some project website. A wiki map for each town keeps track of pictures taken and spots that are still not covered. The goal is to cover as much ground as possible, from a sufficiently large number of angles. The setup can add game type features that activate people more: who will contribute most in our town? Which town will cover most ground? Add some prizes, generate some buzz and we have a hit!
Trouble is, most pictures are taken from a similar location, and almost all in the same horizontal plane (explain the few taken from high up?). So the data set is pretty restricted – notice the hidden areas in the samples.
Hook this thing up to a few xbox connects, some on the ground and a few flying in helicopters and then you have something!!
This is excellent. If you wanted to model a building you could just arrange a flashmob to circle it and take a 100 pics each and bam, instant model.
Combine this with Charlie Stross’ idea that in the near future a large percentage of the population will be continually carrying around cheap mini cameras + data storage devices that record basically every aspect of their lives, and you could have an interesting germ of an idea for a sci-fi story. I’m also reminded of the old 1991 futurist book Mirror Worlds…
Let me go into smart ass mode for a short while:
The Brandenburg Gate, the Berlin Cathedral (or Berlin Dome, rather) and the Ishtar Gate are located in Berlin, Germany, not so much in Rome, with the Ishtar Gate being displayed in Berlin’s Pergamon Museum.
Nonetheless pretty impressiv.
Disabling smart ass mode now…
This should be implemented into google maps
This is nice. But check out Visualsize Inc. http://www.visualsize.com. Their 3D modeling technologies can generate 3D models of a large variety of objects using very few photographs (~ 10 to 20) in a few minutes on a commodity PC – all without camera calibration, special equipment and lighting, man-made markers, and user training. They show that their technologies significantly outperform Project Photofly from Autodesk and Bundler from Microsoft/University of Washington.
graphics rome Technology
Submit a tip
The rules you agree to by using this website.
Who will be eaten first?
Jason Weisberger, Publisher
Ken Snider, Sysadmin