Reading this ("ORB: An efficient alternative to SIFT or SURF") paper, I am not certain how they use the scale pyramid.

It says it is created and FAST feature detectors are employed at every level. Then what? What do they do with these new keypoints found on each level of the pyramid?

  • $\begingroup$ I'm fairly certain they use non-maximum suppression to deduplicate keypoints. As for how they actually combine descriptors over various levels, I have another question which hopefully will get some attention. $\endgroup$ Commented Mar 31, 2021 at 8:28


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.