I discovered IVT when I was searching for a fast and free implementation of Lowes' SIFT features, and I have to say that I'm very impressed with IVTs performance!
I was just wondering if an implementation of SURF features is being planned for future releases. I think it would be great to have a SIFT alternative integrated in the toolkit. I also guess matching SURF vectors would be reasonably faster than matching SIFT vectors.
greetings from austria and thanks for this great toolkit,
david
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Having a SURF implementation would be very interesting. I can't tell when I will have time to implement the SURF features, so anybody who wants to contribute some code is very welcome!
We ourselves use somewhat different features, combining Harris interest points and the SIFT descriptor, and achieving scale-invariance without performing a time consuming scale space analysis by sampling the features at predefined scales. This will be published on the IROS 2009 (International Conference on Robots and Systems), and once the conference is over, I will upload the code. With those features, feature computation takes about 20 ms on current hardware.
And sure, the standard version of the SURF descriptor has 64 dimensions, so matching should become twice as fast...
Cheers,
Pedram
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
hi,
I discovered IVT when I was searching for a fast and free implementation of Lowes' SIFT features, and I have to say that I'm very impressed with IVTs performance!
I was just wondering if an implementation of SURF features is being planned for future releases. I think it would be great to have a SIFT alternative integrated in the toolkit. I also guess matching SURF vectors would be reasonably faster than matching SIFT vectors.
greetings from austria and thanks for this great toolkit,
david
Hi David,
thanks for the positive feedback!
Having a SURF implementation would be very interesting. I can't tell when I will have time to implement the SURF features, so anybody who wants to contribute some code is very welcome!
We ourselves use somewhat different features, combining Harris interest points and the SIFT descriptor, and achieving scale-invariance without performing a time consuming scale space analysis by sampling the features at predefined scales. This will be published on the IROS 2009 (International Conference on Robots and Systems), and once the conference is over, I will upload the code. With those features, feature computation takes about 20 ms on current hardware.
And sure, the standard version of the SURF descriptor has 64 dimensions, so matching should become twice as fast...
Cheers,
Pedram