Great, thanks for the pointer. But... aargh... it's a C# library. What on earth am I supposed to do with that as a researcher? Hmmm... but it did teach me something by pointing to David Lowe's Autostitch. That looks interesting for my other life in science outreach since we've got a backlog of panos from Argentina to process...
Grow up, kiddo. Polishing up innovations is just as important as coming up with them. How many people on/. have ever really come up with a truly original idea anyway?
I'd try to make up a list of 10 things, but I'm not sure what the poster has in mind. Perhaps someone familiar with what goes on at Microsoft, especially Microsoft Research, can come up with one.
Can you throw in a link to the paper and library? I might be able to use this for some non-robotics stuff. SIFT isn't a unique identifier when Googling the pattern recognition literature. And I'd like to see how invariant they are (most algorithms that claim to be scale invariant are only so up to a point) and how they got it to be real-time. (That bit's pretty impressive.)
Have you tried Partiview? It's pretty good with those kinds of datasets - I use it for 3d scatterplots a lot. And you can change the colors of points by whatever attribute you want, and spin the data around smoothly in realtime - I can do that with a quarter of a million points on my laptop with a GeForce Go5200, and probably more if I tried.
Also check out Partiview , a fast open source viewer from the NCSA. It has its limitations, but can handle large animated multidimensional data. The AMNH uses it as the viewer for its nifty planetarium-on-your-laptop Digital Universe, and we've used it to visualize cosmic ray showers, dark matter simulations, global computer networks, clustering patterns, and as a general infovis tool. It also works in stereo, which is really nice when you have two projectors and polarizing filters. The linux binaries supplied sometimes have shared library problems... but/.ers are used to compiling from source anyway...
Slashdot Mirror
Great, thanks for the pointer. But... aargh... it's a C# library. What on earth am I supposed to do with that as a researcher? Hmmm... but it did teach me something by pointing to David Lowe's Autostitch. That looks interesting for my other life in science outreach since we've got a backlog of panos from Argentina to process...
Grow up, kiddo. Polishing up innovations is just as important as coming up with them. How many people on /. have ever really come up with a truly original idea anyway?
I'd try to make up a list of 10 things, but I'm not sure what the poster has in mind. Perhaps someone familiar with what goes on at Microsoft, especially Microsoft Research, can come up with one.
Can you throw in a link to the paper and library? I might be able to use this for some non-robotics stuff. SIFT isn't a unique identifier when Googling the pattern recognition literature. And I'd like to see how invariant they are (most algorithms that claim to be scale invariant are only so up to a point) and how they got it to be real-time. (That bit's pretty impressive.)
Actually, this conjecture was superseded by the Feynman-Tolkien Conjecture that said there was only one ring...
Whoops, add a "/partiview" to the end of that url. Like this.
Have you tried Partiview? It's pretty good with those kinds of datasets - I use it for 3d scatterplots a lot. And you can change the colors of points by whatever attribute you want, and spin the data around smoothly in realtime - I can do that with a quarter of a million points on my laptop with a GeForce Go5200, and probably more if I tried.
Also check out Partiview , a fast open source viewer from the NCSA. It has its limitations, but can handle large animated multidimensional data. The AMNH uses it as the viewer for its nifty planetarium-on-your-laptop Digital Universe, and we've used it to visualize cosmic ray showers, dark matter simulations, global computer networks, clustering patterns, and as a general infovis tool. It also works in stereo, which is really nice when you have two projectors and polarizing filters. The linux binaries supplied sometimes have shared library problems... but /.ers are used to compiling from source anyway...