Closed DrSlink closed 4 years ago
Thank you for being patient, @DrSlink. I revised the PR as follows:
BatchNormalization
parameters,_se_block
,ZeroPadding2D
before the convolutions with strides=2 (not only for TF but also for other backends),Reshape
for 'channels_first',Thank you for the PR!
This is a keras implementation of MobileNetV3. I tried to write this as close as possible to the official tensorflow implementation and the keras implementation of MobileNetV2. This file enables access to the 6 trained MobileNetV3 models found here.