Text this: Visible-infrared person re-identification with region-based augmentation and cross modality attention