Determining the relative pose of an object between two images is pivotal to
the success of generalizable object pose estimation. Existing approaches
typically approximate the continuous pose representation with a large number of
discrete pose hypotheses, which incurs a computationally