How can one pixel of the image correspond to a feature point and a descriptor?

naver / r2d2

Other

461 stars 86 forks source link

How can one pixel of the image correspond to a feature point and a descriptor? #6

Closed ahong007007 closed 4 years ago

ahong007007 commented 4 years ago

Thank you for your wonderful work.

In the paper, a feature point and a descriptor are extracted for each image patch. How can one pixel of the image correspond to a feature point and a descriptor? Thank you.

humenbergerm commented 4 years ago

Hi!

Repeatability, reliability, and descriptor are computed for each pixel. Then we apply non-maximum suppression on repeatability and reliability and keep the top-n (highest rep*rel score) pixel locations as keypoints (+descriptor).

Best, Martin