MobileNetV2:Inverted Residuals and Linear Bottlenecks
论文阅读笔记论文链接(CVPR2018)https://arxiv.org/abs/1801.04381Forexample,thedepthwiselayermayworkonatensorwith144channels,whichtheprojectionlayerwillthenshrinkdowntoonly24channels.Thiskindoflayerisalsocalledabo