One of the Most Impressive IT Things I've Seen!

[sa_20v]

Your thoughts?

http://www.ted.com/index.php/talks/view/id/129

[Graykiraa]

saw this last week and was v.impressed

http://labs.live.com/photosynth/view.html?collecti...

link to the tech demo so you can have a play

[FourWheelDrift]

They used the Guardian to show a good example of news text?

I guess it has a automatic spell check built in.

[alextgreen]

FourWheelDrift said:

They used the Guardian to show a good example of news text?

I guess it has a automatic spell check built in.

...although you'd think an image search for 'boocks' would bring up something a little different

[darkmonsta]

Wow.

Just.... Wow.

[kiwisr]

Surely not that much different from the way Google Earth/Maps has/is been heading?

[off_again]

Seriously cool. I am certainly VERY impressed.

wouldnt mind his laptop that he used though, thats got to be a serious piece of kit!

[CommanderJameson]

me sideways, that's impressive.

[Monte_Cristo]

I'm sure Microsoft has something simliar in development.

[Goochie]

Monte_Cristo said:

I'm sure Microsoft has something simliar in development.

This will be sold as a Microsoft branded product

I have a link on another forum that goes to the M/S site - sadly the forum is down at the moment though.

[mackie1]

Very very impressive indeed. No doubt this will find its way into Virtual Earth (he actually said it would in the demo) and other online apps like flickr.

I'm guessing it takes quite a lot of juice to compute all the relationships between the pictures though.

The seemless zooming is very clever, there must be some clever optimised storage going on to let it stream so quickly.

[kiwisr]

Of course this means that all digital cameras are rendered obsolete, expect the next generation to be supplied with GPS chipsets that encode the location. Until this occurs the creation of things like MS demonstrated will be awfully labour intensitive.

[ThePassenger]

mackie1 said:

Very very impressive indeed. No doubt this will find its way into Virtual Earth (he actually said it would in the demo) and other online apps like flickr.

Video stuttered like bugger here But very neat. You can do similar things already in beryl (with your desktop) but I quite like this.

mackie1 said:

I'm guessing it takes quite a lot of juice to compute all the relationships between the pictures though.

Not really, I suspect in a flickr type situation it'll be running through the tags used. You could even do that with a bigger product like Virtual Earth, just assign more keywords/tags to the image and have it sift through depending on what's selected.

mackie1 said:

The seemless zooming is very clever, there must be some clever optimised storage going on to let it stream so quickly.

It's impressive but not 'that' clever. Modern computers (even, what looks like the stinkpad in the demo) are exceptionally grunty, more so than people realise. The whole thing looks to be done using the 3D accelerator functions so it's a very large texture, as you zoom in it unloads the others or the montage its generated in favour of the image your exploring. I'd expect a pretty impressive chunk of code buffering the image though.

[mackie1]

kiwisr said:

Of course this means that all digital cameras are rendered obsolete, expect the next generation to be supplied with GPS chipsets that encode the location. Until this occurs the creation of things like MS demonstrated will be awfully labour intensitive.

Nope, the whole point was that it's an algorithm working it all out, that's what's really clever!

Obviously any additional information about where a particular shot was taken will makes it's life easier.

[FourWheelDrift]

mackie1 said:

I'm guessing it takes quite a lot of juice to compute all the relationships between the pictures though.

The demo galleries on the website were running here as smoothly as the demo on the video clip and I don't have a monster PC. It's a 2Ghz with 1GB RAM and a 1MB countryside broadband connection.

[kiwisr]

mackie1 said:

kiwisr said:

Of course this means that all digital cameras are rendered obsolete, expect the next generation to be supplied with GPS chipsets that encode the location. Until this occurs the creation of things like MS demonstrated will be awfully labour intensitive.

Nope, the whole point was that it's an algorithm working it all out, that's what's really clever!

Obviously any additional information about where a particular shot was taken will makes it's life easier.

How does that work then? Are you saying I can take a photo of anything in the world and this system will magically place my photo at the right location? Of course not, it's using metadata from the photos.

[mackie1]

ThePassenger said:

mackie1 said:

Very very impressive indeed. No doubt this will find its way into Virtual Earth (he actually said it would in the demo) and other online apps like flickr.

Video stuttered like bugger here But very neat. You can do similar things already in beryl (with your desktop) but I quite like this.

mackie1 said:

I'm guessing it takes quite a lot of juice to compute all the relationships between the pictures though.

Not really, I suspect in a flickr type situation it'll be running through the tags used. You could even do that with a bigger product like Virtual Earth, just assign more keywords/tags to the image and have it sift through depending on what's selected.

mackie1 said:

The seemless zooming is very clever, there must be some clever optimised storage going on to let it stream so quickly.

It's impressive but not 'that' clever. Modern computers (even, what looks like the stinkpad in the demo) are exceptionally grunty, more so than people realise. The whole thing looks to be done using the 3D accelerator functions so it's a very large texture, as you zoom in it unloads the others or the montage its generated in favour of the image your exploring. I'd expect a pretty impressive chunk of code buffering the image though.

I'm well aware of how grunty computers are but you've still got a big bottle neck when it comes to storage. Even with 2+gb of ram being able to zoom in and out of 300 million pixel images like that is very impressive. It'd need to be stored on disk in such a way that you could get the chunk you're after in the resolution you're after in O(1) time. That 300 mega pixel image equates to nearly 860 gig of uncompressed pixel data. Clever streaming code indeed

[mackie1]

FourWheelDrift said:

mackie1 said:

I'm guessing it takes quite a lot of juice to compute all the relationships between the pictures though.

The demo galleries on the website were running here as smoothly as the demo on the video clip and I don't have a monster PC. It's a 2Ghz with 1GB RAM and a 1MB countryside broadband connection.

No I meant to generate in the first place, not to view the collections.

[King Herald]

So, a bunch of digital pictures in extreme detail clipped together using latest digital technology


What did I miss?

Some sort of geekfest???

[mackie1]

kiwisr said:

mackie1 said:

kiwisr said:

Of course this means that all digital cameras are rendered obsolete, expect the next generation to be supplied with GPS chipsets that encode the location. Until this occurs the creation of things like MS demonstrated will be awfully labour intensitive.

Nope, the whole point was that it's an algorithm working it all out, that's what's really clever!

Obviously any additional information about where a particular shot was taken will makes it's life easier.

How does that work then? Are you saying I can take a photo of anything in the world and this system will magically place my photo at the right location? Of course not, it's using metadata from the photos.

The algorithm will need to be fed a collection of images of the same thing but once it has that it compares all the pictures in the collection and builds the 3d picture based on image content, not positional information. Although I'm sure it could be made to take advantage of such information if it were available. So yes it would need some metadata in a flickr-like environment but after that the process is automagic.

At least that's how I understand it. If the demo was created manually then it's just not very impressive.