Rethinking Rotated Object Detection with Gaussian Wasserstein Distance Loss

         Xue Yang 1 2 3 Junchi Yan 1 2 Qi Ming 4 Wentao Wang 1 Xiaopeng Zhang 3 Qi Tian 3

             Abstract
    Boundary discontinuity and its inconsistency to
    the final detection metric have been the bottle-
    neck for rotating detection regression loss design.
    In this paper, we propose a novel regression loss
    based on Gaussian Wasserstein distance as a fun-
    damental approach to solve the problem ...