Geek Answers Handbook

Are there any detectors implemented on the GPU and having a scale / rotation-invariant?

As is known in OpenCV 2.4.9.0, these detector functions are: SIFT, SURF, BRISK, FREAK, STAR, FAST, ORB .

They all have an implementation on the CPU, but only FAST and ORB on the GPU. http://docs.opencv.org/genindex.html

And as you know, some of them are scaled / rotated-invariant, but some of them are not: Are there any quick alternatives to SURF and SIFT for extracting scale-invariant features?

They are scale invariant and rotate-invariant:

  • SIFT
  • Surf
  • Brisk
  • THE GAME
  • STAR

But they are not scale-invariant and do not rotate-invariant:

  • FAST
  • ORB

Are there any detectors implemented on the GPU and having a scale / rotation-invariant?

Or will it be added in OpenCV 3.0 on the GPU or OpenCL?

+8
opencv computer-vision feature-extraction feature-detection orb
source share
2 answers

Actually, SURF is the only feature detector with zoom / rotate function with GPU support in OpenCV.

OpenCV 3.0 FAST and ORB have OCL support, and in addition, the two ( FAST and ORB ) have already received CUDA support.

Support for OCL / CUDA SURF has already been mentioned in the comments on your question, but this is only a contribution to OpenCV, and that is exactly what OpenCV developers of opencv_contrib do :

New modules quite often do not have a stable API, and they are not well tested. Thus, they should not be released as part of the official OpenCV, as the library supports binary compatibility, and tries to provide decent performance and stability.

Based on my previous experience with OpenCVs, the implementation of SURF functions was much weaker than OpenSURF . It would be wise to try this or find some other open source versions.

ps: As far as I know, there is no accelerated version of the GPU KAZE / AKAZE .

+8
source share

I recently implemented AKAZE using CUDA with several colleagues, if you are familiar with the original library, you should not have problems with its use, since we respected the API. The current version can be found here:

https://github.com/nbergst/akaze

+3
source share

More articles:


All Articles