why did orb features size not decrease same times when I reduce the image's size?

Question

everyone! As I use ORB feature extraction, my image is 640*480, and orb descriptors Mat is 500 rows and 32 cols, I think when I reduce the image size to 320*240 by using opencv::resize() function, the image orb descriptors will reduce to be 250 rows and 32 cols. However the answer is 420(different image will be different, maybe 417, 430) rows and 32cols. Can anyone tell me why is that?

Answer 1

Nice questions. There is a default on the number of points detected by Orb to be 500.

In the first scenario, it probably detect a lot more then 500 points, but retains the best 500.