New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Batch norm in Mobilenet v2. #3

Open

meet-minimalist opened this issue Jul 17, 2020 · 0 comments

Owner

meet-minimalist commented Jul 17, 2020

As per the architecture definition provided here, it is shown that the batch normalization is used in inverted-residual blocks as below.

Bottleneck layer which is normal convolutional layer is equipped with batch norm
Depthwise layer is also equipped with batch norm
Pointwise layer is also equipped with batch norm.
Even in their paper they mentioned that they use batch norm after every layer.

But when we download a pretrained model from tensorflow and visualize it in Netron, it is as per below.

Bottleneck layer doesnt use batch norm instead use bias.
Depthwise is equipped with batch norm.
Pointwise layer doesnt use batch norm instead use bias.

So this make a huge difference in number of parameters and final accuracy.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment