Can't get bias from checkpoint because of using batch_normalization when training

[chenping5121]

I want to get bias and weights of every layer from Checkpoint, but I can't get bias instead of getting batch_norm_moving_average, batch_norm_moving_variance and batch_norm_moving_beta, So I can't use these data to test my own images. That's means if I don't use batch_nomalization, I can get both weights and bias of every layer? And I use your Pretrained Model to generate the binary bias and weights, I can get, which I am confused? How you train the model to get Pretrained Model ?
The binaries of your Pretrained Model Checkpoint and my training model are as following：