SIFT implementation on TX2

Hi all,

I am wondering if there exists an implementation of SIFT (including FLANN+RANSAC for matching) for TX1/2 . I am currently using the one in OPENCV 3.2 but it is CPU based so slow. I am aware that there are CUDASIFT or GPUSIFT online but still want to know if any more formal version exists in visionworks or tegra-opencv.

Thanks!
Ray

Hi,

We have Harris and Fast implementation in Visionworks.

Please download the library and sample code via JetPack.
Please find nvx_demo_feature_tracker for more information.

Thanks AastaLLL!

In mind, those two are more suitable for tracking (like SLAM).
May I ask if NVIDIA has a plan of including SIFT into the Visionworks?

Hi,

Thanks for your question

We currently doesn’t have plan for SIFT.
We consider performance, so we implement features extraction function with harris and fast.

Thanks.