1 paper · 1 filter
Kohei Yamashita, Vincent Lepetit, Ko Nishino
Computer vision has long relied on two kinds of correspondences: pixel correspondences in images and 3D correspondences on object surfaces. Is there another kind, and if there is,…