TF 2.x Reading parameters from keras BN Op

General Discussion
, ,
#1

Hi,

I am new to TF 2.x. When BatchNormalization layer is created in TensorFlow 2.x, am unable to read BN parameters (such as momentum, epsilon etc.) the way we used to in TF1.x. Is there a way to work around this? Specifically, if we have Keras BN op with no training argument, how do we read the associated parameters such as beta, gamma, epsilon, momentum , moving mean and average from it?

#3

Hi @codingrat, The image related to your post is in form of Preformatted Text. Kindly upload the image for better solutions.