no code implementations • CVPR 2017 • Di Xie, Jiang Xiong, ShiLiang Pu
Moreover, we can successfully train plain CNNs to match the performance of the residual counterparts.