Convert an image into non-overlapping Vision Transformer patch embeddings with a learned linear projection.
