no code implementations • CVPR 2017 • Quanquan Li, Shengying Jin, Junjie Yan
More specifically, we conduct a mimic method for the features sampled from the entire feature map and use a transform layer to map features from the small network onto the same dimension of the large network.