Deep learning method that trains two networks simultaneously without extensively annotated training data that compete with each other in a zero-sum game framework.