no code implementations • 22 Sep 2018 • Huasong Zhong, Xianggen Liu, Yihui He, Yuchun Ma
These three primitives (channel shift, address shift, shortcut shift) can reduce the inference time on GPU while maintains the prediction accuracy.