Learning to Recognize Objects from Unseen Modalities: Supplementary Results