Some details about the pre-trained model

[liu-zhenhua]

Hi, thanks for your code.
I have two confusions about the pre-trained model.

1. The order of input channels is RGB or BGR?
2. I see the dropout_rate is set to None in the code and did you use dropout layer in the pre-trained model?
   Thanks again and look forward to your apply!