chunked/nltk_chunking/model_doc_clip/chunk_19.txt

To feed images to the Transformer encoder, each image is split into a sequence of fixed-size non-overlapping patches,
which are then linearly embedded.