Microsoft is working on Microsoft Live Labs photosynth, which sounded like a version of AutoStitch on steroids, but is actually something else entirely.
Basically, what Photosynth tries to do is place a photo you’ve taken in the exact world by mapping the structures in the data against the rest of collections of your images supplemented with a database they’ve created from geopositioning data.
It then will allow you to fly through a virtual world. of these pictures placed in space and then you can select the interesting ones as you navigate and view it.
The issue of real landscape photography
The first issue is that if you look at how the navigation system is rendered, all the images show 2D texturemaps of the images in a 3D space. This implies that photos with a lot of depth at multiple points in the image wouldn’t have any representation here. In other words, the only photos it can deal with are detail shots and wide angle landscapes.
I’m curious how the system might deal with a classic landscape photographer’s shot. In this sort of shot, an ultra-wide angle lens is used and the horizon is in the top half in order to have something with foreground interest carry the viewer into the background.
Theoretically, this could be dealt with by having photosynth map the 2D image surface at an angle in 3D space, so that the foreground is really close to you and the background is really far away. I have a feeling it will just die at rendering it in the proper place in Photosynth and I’d probably have to train it to only try to map the Fort McDowell barracks in the background.
The issue of natural features
The second issue I have is with natural features. Basically the computer vision algorithms used are only demoed against famous landmarks like St. Marks Basilica in Venice. Granted the last time I was in St. Marks, I was only ten-years old, but from what I remember, it has got to be the most feature-rich area for a computer vision system.
I wonder how it will perform at placing all my photos of the Golden Gate Bridge?
And how about natural shots? Would it be able to distinguish between various plunge waterfalls?
Does it use EXIF and GPS data to assist in placing this photo?
This is a popular photography point, I wonder if it can place it…
A comment about the advertising
Microsoft is becoming heavily Apple-influenced. The production of the video is obviously a homage on the Apple product videos.
Note to Live Labs: Apple has moved on. Please check out their Aperture Profiles (click on Heinz Kluetmeier, Joe Buissink, or Richard Burbridge to see videos) or their recently released Final Cut Studio Profiles. Watch. Learn. Steal.
BTW, I love your PhotoSynth logo. The idea of using leafs as an aperture to play off the word: “photosynthesis”. Very clever and definitely un-Apple. Kudos to the logo designer.
The take home point
I love Microsoft Labs idea here. It’s nice to see them trying to bite off more than they can chew.
But it’s clearly geared at your typical travel photograph—I suppose the most interesting thing I could gather here is watch where everyone photographed themselves holding up the Leaning Tower of Pisa.
When Nikon invented the first evaluative metering system (“matrix meter”), they analyzed the exposure data from thousands of professional photographs. Now over 300,000 of the best photographs go into the creation of the 3D color matrix meter in the latest Nikons.
A lot of us trust the Nikon system, even though we don’t go take photographs with the same equipment, with the same lens, at the same time of year, from the same coordinates that Ansel Adams did (as verified by a GPS).
My point? How popular would it have been if Nikon used people’s random snapshots in creating their metering system? Would we have bought into that technology?
What photos are we really interested in?
What photos does Microsoft use to build their recognition system? Photos taken by professionals who try to show us what we’ve seen before in a novel way, or ones taken by tourists?
That’s my problem with PhotoSynth.