Subscribe

Follow Me

Now Reading

  • Soldier Spies (Men at War)

    Soldier Spies (Men at War) by W. E. B. Griffin

Recently Read

  • Coup d’Etat

    Coup d’Etat by Ben Coes

  • A Passion to Win

    A Passion to Win by Sumner Redstone

  • The Zero Hour

    The Zero Hour by Joseph Finder

  • Engines of Change: A History of the American Dream in Fifteen Cars

    Engines of Change: A History of the American Dream in Fifteen Cars by Paul Ingrassia

  • The Last Refuge: A Dewey Andreas Novel

    The Last Refuge: A Dewey Andreas Novel by Ben Coes

See Full Library

Checkin’ Out Photosynth

When I saw the coming out demo of Microsoft’s Photosynth technology done at the TED Conference over a year ago, I was totally blown away.

Photosynth automatically assembles a set of individual photos of a particular subject into a three dimensional, explorable universe of the scene. The more photos, the more detail and the more explorable the final “synth.”  It differs from stitching – the process of aligning and joining several overlapping photos to create a single larger image – in that the resulting image is a space rather than a flat 2D image.

When the public beta was introduced a couple of months ago, I was all over it.  I played with other people’s synths and was impressed.  But, of course, I had to give it a go myself.  I decided to throw what I thought would be a difficult scene at it – one with trees.  Trees always give stitching programs fits and, as it turns out, they do the same for Photosynth.  There are just a whole lot of edges to align.

I took 144 photos of a location (you don’t need to take that many, but I wanted to see how complete a scene I could create) from every angle I could get to.  Photosynth cranked on the photos for a while and broke the scene into many different views.  There should have only been one, but the program couldn’t match up the views to form a single synth.  The results are below.

Photosynth reported that my 144 photos were only 23% “synthy.”  Basically, Photosynth could only make heads or tails of 23% of my photos in creating the final synth.  If you look at the synths on the web site, you’ll find excellent ones that are >90% synthy like The Boxer.

The user interface for creating synths is very simple and the program creates synths with virtually no user intervention.  Exploring synths is a different matter.  The browser interface is a bit strange to me.  I’m never sure what the arrows and buttons are supposed to do, even after trying them.  I may be using it to its fullest, but I may be missing the point entirely.  A few more tooltips might be helpful.

You need to download the Synther, which runs on your PC (no Mac support yet).  The Synther will upload the synth to servers in the cloud.  You’ll need a Microsoft Live ID to use the service.  For now, all uploaded synths are public.  Everything is free.

I think this technology has tremendous promise and I plan on playing with it a lot more.  Of course, I’ll report back on my findings.  In the mean time, you may want to give it a try.  It’s easy and very cool.

Reblog this post [with Zemanta]
Technorati Tags: ,
  • john bower

    Will,

    So I first saw this on DPReview and gave it a quick look, then I believe you mentioned once to me.

    I think some of the navigational confusion comes from thinking this is a type of photogrammetric program. One that remaps the pixel information onto 3D surfaces and results in essentially a single texture mapped 3D surface file.

    In fact this appears to something more and less than that. In essence, the program appears to create a series of hyperlink tags that contain the 3D location and orientation of each of the images in a synth group. The synth viewer is the interface through which one navigates these hyperlinked files.

    Instead of an underlined set of text to indicate a hyperlink, a ghostly trapezoid outline appears indicating another image file that shares common spatial information with the image that is front and center. If another image is a complete subset of the present image, e.g. a close-up view, you can access that information by either clicking within the outline or merely zooming in.

    What is quite fascinating is its paradigm for categorizing and accessing huge amounts of visual data. That and the essentially automatic process to do the categorizing.

    This is also the first significant use I’ve seen of the jpeg2000 file format. Which also shows its great potential.

    John

  • john bower

    Will,

    So I first saw this on DPReview and gave it a quick look, then I believe you mentioned once to me.

    I think some of the navigational confusion comes from thinking this is a type of photogrammetric program. One that remaps the pixel information onto 3D surfaces and results in essentially a single texture mapped 3D surface file.

    In fact this appears to something more and less than that. In essence, the program appears to create a series of hyperlink tags that contain the 3D location and orientation of each of the images in a synth group. The synth viewer is the interface through which one navigates these hyperlinked files.

    Instead of an underlined set of text to indicate a hyperlink, a ghostly trapezoid outline appears indicating another image file that shares common spatial information with the image that is front and center. If another image is a complete subset of the present image, e.g. a close-up view, you can access that information by either clicking within the outline or merely zooming in.

    What is quite fascinating is its paradigm for categorizing and accessing huge amounts of visual data. That and the essentially automatic process to do the categorizing.

    This is also the first significant use I’ve seen of the jpeg2000 file format. Which also shows its great potential.

    John

  • http://www.2-speed.com/ Will

    John,

    That’s a great description of how the UI works and that’s how I understand it, too. I guess my problem with the UI is not so much that I don’t understand it, but more that I think it could be better or, at least, more intuitive. I can see it turning many people off.

    I agree that there’s tremendous power here. I just think that Microsoft could make it more accessible or, at least, less intimidating.

    I didn’t realize that they were using JPEG2000. Cool.

  • http://www.2-speed.com Will

    John,

    That’s a great description of how the UI works and that’s how I understand it, too. I guess my problem with the UI is not so much that I don’t understand it, but more that I think it could be better or, at least, more intuitive. I can see it turning many people off.

    I agree that there’s tremendous power here. I just think that Microsoft could make it more accessible or, at least, less intimidating.

    I didn’t realize that they were using JPEG2000. Cool.