Text this: Learning enhancing modality-invariant features for visible-infrared person re-identification.