Microsoft Demos "Deep Zoom" Technology

Barence writes "Yesterday, during a presentation for this year's Imagine Cup, Microsoft's Mark Taylor demonstrated the company's Deep Zoom technology to appreciative gasps of admiration from the computing students present. It's pretty impressive stuff, and you can try 'deep zooming' for yourself at the Hard Rock Memorabilia Site." Unfortunately the demo requires the Silverlight plugin and the story is pretty thin on technical details. I would be interested to see how they captured the image data to that level without massive pixelation.

Haven't you ever..

seen CSI? This technology is so passe.

No, but I've seen GigaPan

I don't watch much TV, but the functionality is awfully similar to GigaPan [gigapan.org].

DeepZoom

My understanding is that you use different resolutions of the photo. The original photo is obviously the highest res you can have, but you can make successively lower res copies. More or less just bring up a a higher res version when the user clicks.

I saw this demoed at the Atlanta Code Camp [atlantacodecamp.com] back in March. Very cool to watch.

oh lordy...

I would be interested to see how they captured the image data to that level without massive pixelation.

You don't ... you don't actually think that the image data came from one photo ... do you?

*slaps forehead*

Unfortunately?

Unfortunately the demo requires the Silverlight plugin...

A Microsoft tech demo requires the installation of new Microsoft software to view? Who would have though?

While Silverlight might never be as widely-supported as Flash, I hope that perhaps the competition might force Adobe to do something about the CPU hog that is Flash.

Maybe not CSI

But how is this different different from google maps (or live maps, or WHATEVER allows you to zoom out a lot)..

Re:Maybe not CSI

A major difference is the inclusion high resolution collections, which are not fixed at runtime and can be rearranged programmatically. I know this because that is what we did on the Hard Rock Memorabilia project.

Aside from that, it is another form of a "tile server" application... Just one that happens to be rather easy to use from a development perspective, and one that has been done really well (Zoomify/AJAX-based solutions don't hold a candle to the tile stitching and easing effects built into the MultiScaleImage control, IMHO).

SeaDragon

Silverlight's MultiScaleImage control (aka deep zoom) is a version of the SeaDragon renderer. The image format it uses is a custom tree structure that contains pixel details relevant to both it's position in the tree and relative to it's peers. Essentially, it's a hierarchical image with very smooth transitions.

Silverlight: silverlight.net
SeaDragon: http://labs.live.com/seadragon.aspx [live.com]

layered bitmaps

The Beatles models and signatures pear to be the highest level of detail unless there are other "Easter eggs". That level of zoom on any surrounding areas is pixelated. They have stacked multiple high res photos at various scales in this particular area.

Yet another Deep Zoom

Ian Griffiths [interact-sw.co.uk] implemented a deep zoom for the BBC [bbc.co.uk] in their Big Weekend festival. Rather pleasingly they chose to call it the "Big Zoomy Thing" in a nice bit of anti-jargon.

Youtube video of similar demo

Here's a Youtube video of a very similar demo from a past keynote, no Silverlight required:
http://www.youtube.com/watch?v=xa-2-eYuJuk [youtube.com]

It's based on the same Hard Rock Memorabilia website, but shows slightly different sections of it.

WTF?

If you're like me and a bunch of very smart students, you can't fail to be impressed.

I must be dumb.... Stiching together an image of higher-res photos might be a technical wow, but sorry, I'm not really impressed. This sort of thing I might expect from a college lab, but for a multi-billion dollar company to present this as some sort of earth-shaking innovation?

Uses gigapixel imagery as source

There is a bit of a misdirection in articles and other material about Deep Zoom.

Most people go ooh and aah because they (wrongly) assume that it zooms into normal resolution photos .

It doesnt (because as you and I know, it physically can't).

Deep Zoom does NOT perform CSI/CIA-style photo enhancement. If you dig deeper, you will find that what Deep Zoom is intended for is to enable one to focus on a smaller portion of a giga-pixel photograph so you do not have to download the whole photograph.

Think of it like a hierarchical smooth slicing of a large high resolution photograph and only downloading those "planes" and "sections within a plane" that the user is interested in seeing.

Interesting technology but not magic.

Re:Uses gigapixel imagery as source

They actually use SeaDragon (the name of the technology) on CSI, for those sections you're talking about. Obviously they lie about what it's doing, but that's the software you see.

No free lunch

There is two ways to get this level of zoom to work:

1) have the pixels in the first place
2) having more pixels in the first place.

Anything else is a fundamental violation of the laws of physics and math. You simply can not fake what you don't have without it being exactly that: a fake. There is no storage printing technology which could accomplish this level of zooming, and they carefully do not say that this is actually a continuous zoom of a picture on a stamp.

Deep Zoom works by letting you meld several images in such a way as pretend its one image.

Basically, its a con-job of transitioning several different images, where one is a re-photograph of sub portion of the original.

The implication of the article is that this is all one image containing a nearly infinite level of detail, which it most emphatically is NOT.

The author is probably equally impressed by street corner magic tricks.

Re:No free lunch

Deep Zoom works by letting you meld several images in such a way as pretend its one image.

That's still very useful.

Basically, its a con job of transitioning several different images, where one is a re-photograph of sub portion of the original.

'con job' has needless connotations of an intent to deceive.

The implication of the article is that this is all one image containing a nearly infinite level of detail, which it most emphatically is NOT.

No. The implication of the article is that you can provide this as a user interface, which is very cool. Google Earth isn't interesting because its a 'con job' to let us think we can zoom in and out of a single monster image of the planet. Its interesting because its a natural and convenient UI to use.

And we don't have to download every single pixel of every single higher res image of a tree in Nigeria to have a closeup look at a parking lot in London. Detail is loaded on the fly, as needed, while the user gets a 'seamless' and comparatively low bandwidth experience.

Its not particularly new as an idea. Or even as an implementation. But maybe Microsoft's tools make setting it up substantially easier, and that alone would be a nice bit of progress.

The author is probably equally impressed by street corner magic tricks.

I am impressed by street corner magicians too. Not because I think they're magical, but because I am impressed at their showmanship, sleight of hand, dexterity, and general ability to appear magical.

I don't get it

Don't we already have the ability to process multi-resolution images in, for example, Google Maps? You know, zooming in and out images with large total resolution?

It would be impressive if the photo they demonstrated on was anything but a photoshop, but given that the 428x134 signature is 52x11 in the 350x237 statuette picture which is 29x26 in the 428x350 hard rock picture which is 87x87 in the 428x399 stamp picture, for the stamp to be real would require a 33 gigapixel stamp (which, at 1 inch square, would be printed at 33,000,000,000 DPI).

To me zooming in and displaying a different image isn't really as exciting at the article author makes it sound? Maybe I'm missing something because the journalist sounds pretty damn excited about it.

The image is only 21K

But the viewer is 126G.

Crashed FF 3.0 on my Mac

It crashed Firefox 3.0 on my Mac Book after installing the plug in and viewing the demo.

Re:Crashed FF 3.0 on my Mac

It consistently crashes Firefox 3.0 RC2 on Vista 64, too, after installing the Silverlight plug-in. I disabled the plug-in and no crashes.

Of course the Silverlight and the zooming works as advertised in IE 7.0.6

Re:This is not new...

Typical Slashdot... they post a snarky anti-Microsoft comment with a pretentious air of superiority but get the details wrong.

Photosynth is not Deep Zoomm. Photosynth reconstructs 3D models from collections of 2D photos of the scene acquired from different positions and angles. And as far as I know, Photosynth wasn't an acquisition - it was produced by Microsoft Research.

Deep Zoom was an acquisition, but it was the technology formerly known as Seadragon. It's completely unrelated - Deep Zoom/Seadragon is a 2D thing.

And it's an acquisition, but so what? Ooh, naughty Microsoft - how dare they take exciting technology developed by a startup and put it in the hands of millions of users? Shocking! Clearly it they should have left it to sink in obscurity.

Re:This is not new...

Photosynth was acquired from the University of Washington... The original was in Java and called photo tourism. http://phototour.cs.washington.edu/ [washington.edu]

Re:they stole it from blade runner and csi

I'm far more interested in the tech that allowed them to change the camera angle of the photographs in Bladerunner. When's that coming?

Re:Installing Silverlight

Sibelius [sibelius.com] is a popular music notation software package.
It has become pretty popular in the past 5ish years since its learning curve isn't nearly as steep as its main competitor Finale.

People criticize Sibelius since, typically (at least for the versions I've used), its output isn't exactly professional quality.
It is, however, a great tool for music students.

Back in the day, Finale was the only option for amateur composers to produce professional looking manuscripts.
I'm not sure how far Sibelius has come in the last few years, so things might have changed.

Re:Installing Silverlight

Similar hurdles exist for indexing silverlight content as they exist with flash. Silverlight is mainly for media and data/info visualization.

It's technically possible to index silverlight 1 content, because it's content is "loose Xaml files", which means the site has xml files alongside html/js/etc, that is rendered by the silverlight 1 engine.

Silverlight 2 has the same capabilities, but noone will use them, because using C# for application/interaction logic is way more productive than using Javascript. Silverlight 2 sites using C# have the following structure

SomeSite.XAP (zip file containing all code and assets)
- AppManifest.xml
- ApplicationCode.dll (.NET Assembly containing Entrypoint and embedded assets)
- SomeResources/ (compressed folder)
- SomeResources/SomeImage.jpg (...)

AdditionalContent.XAP (supplemental resources and code)
- AppManifest.xml
- SupplementalCode.dll

This makes silverlight 2 apps and content updates really easy to, but are a barrier to extract information.

In both cases the information gained isn't nearly as useful as textual html content, and completely different heuristics would be necessary to analyze the importance of one unit of textual content vs another. Indeed, nearly all the visual cues (The relative position, color, highlights, animations, and reactions to the user) would likely be lost in the process. Perhaps the search engine that can index flash and silverlight content is one that analyzes both visual and textual content.