Access current input batch in TensorFlow 2.0+

Question

I want to write a loss function that depends on the input pattern, (using TF2.0+)

Looking at this SO thread and other similar threads, I can understand only how can I