DEALIO: Data-Efficient Adversarial Learning for Imitation from Observation