Slashdot Mirror
Bing Maps Wows 'Em At TED2010
theodp writes "In an eye-candy filled presentation that earned him a standing-O at TED2010, Blaise Aguera y Arcas demos augmented-reality mapping technology from Microsoft. In his eight minute spiel, an extension of a shorter tech preview video, the Bing Maps architect shows how geo-tagged Flickr images can be precisely incorporated into streetside views, demonstrates indoor panoramas at Pike Place Market complete with live video overlays, and even takes the audience into space with Microsoft's Worldwide Telescope. " This is a really exciting video and worth your 8 minutes.
"3D is currently not supported for your browser. For a list of supported browsers, see Help."
Seeing help:
Supported browsers.
* Internet Explorer 6 or later
* Mozilla Firefox 3.0 or later
* Safari 3.1 or later
I'm using Firefox 3.6. But I guess it's not my browser that isn't supported. It's probably because I'm running it on Gentoo. I guess I will have to stick with Goggle Maps after all.
[sarcasm] One more point for Microsoft for web neutrality.[/sarcasm]
Plus you have the crowd effect. I've been to plays and concerts where I didn't think it deserved a standing ovation and based on other folks sitting, they didn't either. But a bunch of people stand, then the folks around them stand, and it continues until everyone is on their feet. But it doesn't cost me 20% of the ticket price and it's a good chance to start out for the car :)
[John]
Shit better not happen!
Huh? Did you watch the whole presentation? The flickr images displayed in 3-D in-place in the street view? The LIVE video being overlayed in-place in the street view, following the camera pan in real-time? For that matter what about the smooth zooming in/out of the map itself vs Google Map's stop-and redraw at next level.
Bitch all you want about Microsoft, but it was a very impressive demo. Kudos to the software guys who developed this stuff.
As someone who's being developing software professionally for 30 years I tend to by cynical and blase, but stuff like this really is impressive and makes you stop and say "Wow!".
Never mind years ago, I challenge you to show me just ONE other app today that can, for example:
1) Take a random geo-tagged photo (flikr photos in the demo) and integrate it in 3-D into it's EXACT (not just geo-coordinate) correct spot in a 3-D scene
OR
2) Integrate live video into a 3-D scene following the camera pan in real-time
And, no, Google maps "pin the tail on the donkey" displaying of photos at geo-tagged locations is not even remotely the same thing. An idiot could do that. Microsoft is recognising the map scene in 3-D and (itself an extraorinarily difficult task) correlating that to 3-D adjusted photo content. This isn't an "incremental improvement" unless you consider the space shuttle an incremental improvement to a cart pulled by a donkey.
Here's my million dollar idea. Why can't I have a search engine where I can click on a search result 'never show results from this domain again'. It might take awhile but you could build up nice filtered list after awhile. Hell, even being able to share your list with people and the community builds a good filtered list to get rid of the crap.
Yes, I did.
...
You mean like this?
Then no, you didn't. You said you did, but obviously you didn't. The only slightly similar thing is that in google earth when a user clicks on a link, it will zoom into a position where the image perfectly aligns (if the person who authored the link successfully made it align.) Thats in contrast to what Microsoft is doing where no matter what orientation the user has put himself, the image will be morphed to align, and that no link authoring is necessary at all (nor any tedious positioning, by definition)
You mean that irrelevant eye candy effect that google earth had since it was first released?
Google Earth does not do this with the overlayed images. To get the overlayed images, you must click on a link to them and then the camera is moved to a specific position for viewing. Essentially, this google earth feature is stupidly not useful at all and has simply been hacked into their earth client with the absolute bare minimum of effort.
It makes me wonder if you are aware of the tools which have been available for, say, the past 5 years.
I do not wonder weather or not you viewed the demonstration video. I know you didn't. You couldn't have without being so retarded about whats in it.
"His name was James Damore."
Here's the URL for the video on the TED site, in a larger format, and without "techflash" anywhere nearby:
http://www.ted.com/talks/blaise_aguera.html
When I was a kid, we only had one Darth.
Reposting logged in:
To people interested in image based rendering, something like the system presented by Microsoft is inevitable, yet still impressive when actually implemented. Look at the transitions in Google Streetview, for example: You have to pay close attention because it happens really fast, but you can see that Google also has a 3D proxy underneath the images. The transition is not between different projections of flat images but between rough approximations of the actual geometry, textured with the image data. That is what makes Microsoft's system so seamless as well. The existence of an underlying geometric understanding of the scene is also obvious when you move the cursor over a Streetview image or look at the cursor in the TED demo: It changes perspective depending on the geometry.
The critical algorithm at the core of it all is called "SIFT" (Scale Invariant Feature Transform). That's what enables the computer to identify matching features in different pictures, as long as they're taken from similar positions. (This is done after prefiltering the images according to geo-tagging information to reduce the search space.) Then you have sets of 2D coordinates of 3D points under several projections (images). This data defines a set of equations which you can solve to get the relative camera positions and 3D coordinates of the feature points. If you've followed the news on PhotoSynth, you might remember pictures of 3D point clouds: Those were the calculated 3D positions of feature points in the source images. From these point clouds, you can create an approximate representation of the geometry of the scene. If you then use the picture taken from a position closest to your current viewpoint to texture that geometric proxy, you get what Microsoft presented at TED. It really isn't all that complicated.
Inevitable, therefore not really surprising, but still mighty cool.