After a few days, a mixed verdict on Microsoft Photosynth

It's a new tool from Microsoft that enables photographers to upload photos, but then let other users walk through those photos as though they showed 3D scenes. In some ways, it's close to amazing. Close.

At one level, Microsoft Live Labs' public launch of its Photosynth project is an intriguing test of a commercial software producer's ability to perform supercomputer-style computations as a service for the general public. At another level, it's a clever and somewhat effective scheme for getting more users signed onto Microsoft's Windows Live network, as well as using plentiful amounts of its online storage.

But there are still some lingering questions about the service's practicality and staying power, past its ability to fascinate new users. BetaNews has been putting those questions to the test, in a trial which lasted longer than we initially anticipated.

Photosynth is an online tool that evaluates the content of multiple photographs which contain much of the same subject matter, shot from differing angles or at different times. Its objective is to assemble a three-dimensional montage that pieces together those photos at precisely the parts where they overlap, even if the angles at which those contents were shot changes drastically. It then provides a 3D walkthrough tool for users to peruse each photo as though she were walking (or traveling, or flying) along the same routes the photographer took.

A simple example might include a sculpture photographed circumspectly at 30-degree intervals. Some of Microsoft's more convincing examples include flying over a downtown area, or touring a garden.

As Microsoft Group Product Manager Alex Daley explained to BetaNews, the concept may be most applicable to touring a building, either from the inside or outside. For lack of an original name for things, the tool you use to produce these montages is called a synth-er.

"Generally, we take one shot of this building, one shot of that building, or one shot of the front of a room and one shot of the back of a room. But here, if you take photos of about 40 or 50% overlap to each other -- which is very simple to do in the digital camera era, and it...doesn't what matter what kind of camera, even camera phone -- if you put them into the synth-er, it'll actually calculate what it thinks the environment looks like by looking for similarities between the textures and the object. So even if the lighting conditions are different, even if they were taken at a different time of day, different time of year, different angles, Photosynth can kind of match up the textures and it creates this 3D 'point-cloud,' which you can see floating in the background behind those pictures. It's a very sparse 3D model of the area. Then we use that 3D model as a canvas on which to display the photo."

The 'point cloud' shows up as a field of dots in Photosynth's 3D universe, representing points where facets of objects may appear in one or more photographs. At any one time, you look at the contents of one photo (sometimes a hazy representation of other photos appears in the background), but the points help you to get some kind of bearing for the active photo in the context of the rest of the scene. It's Photosynth's way of attaching a placeholder to where it thinks something should go.

"[Photosynth] relies 100% on the ability of the software to extrapolate a 3D model from the 2D," said Daley. "There's actually two major technical advances in Photosynth that, together, are kind of a 'one plus one equals five' scenario. Number one is the synth-er itself, the software which, when you install the viewer, is also installed on your machine. You can just click on the Create button at the top of the page, click 'Create a Synth' on that page, and it'll actually open a synth-er for you. You just give it a set of 2D photographs -- they have to be a little different than how you'd typically take them, just in the sense that you want to take photographs with a lot of overlap of each other.

"The other half of that two-part achievement is the zooming technology. If you notice, we can load lots of very large, very detailed pictures very fast. As a matter of fact, you can even fill your screen with a Photosynth, there's a small button in the bottom right of the viewer that'll maximize it to the full browser window area, and you'll see we can even fill in pictures with detail, and you can zoom in on any image and get all the details in there. So whether you're talking about 5, 10 megapixel images, or if you're talking about a synth like, you can search for the National Archives and you'll find a synth where we have a bunch of 10 Mp images from a digital camera mixed with 50 Mp scans of the Constitution, and it was able to match them up [and situate them in the scene] automatically -- even though those were taken with a different machine in a different location, the Constitution's still the Constitution."

Nothing about the original photographs particularly changes, Daley pointed out; in other words, you're not just seeing the "merged" content from all photos put together. If there are people or moving objects or birds in one photo that aren't present in any of the others, you'll still see them. "We don't want to change the photo," remarked Daley. "To us, the photo is sacred."

Another Microsoft example takes advantage of this scheme: A mother horse parades her new foal through a paddock. Photosynth stitched together the several angles in which we follow the pair, though as we move from point to point in the field, it's as though we're moving with them.

Next: Not exactly a walk in the park for Photosynth...