In tensorflow v1 api we can predict by passing input to specific intermediate layer as below, how we can do the same in keras
session.run(\'predictions:0\',
