ShuffleNetV2 from the paper: https://arxiv.org/abs/1807.11164.
This is not a official implementation.
System: Ubuntu 16.04, python3.5, tensorflow-1.14.0. tflearn-3.2.0
We use tensroflow and tflearn in a simple way to construct the ShuffleNetV2.
Using the oxford 17 catergories flowers dataset. Because its image size fit the papper. But there is a different part, we found the final output size with convolutionaloperation not like the papper 7x7, is 6x6, so we change a little bit.