chunked/nltk_chunking/model_doc_pix2struct/chunk_8.txt

For the first time, we show that a single pretrained model can achieve state-of-the-art results in six out of nine tasks across four domains: documents, illustrations, user interfaces, and natural images.