chunked/nltk_chunking/model_doc_owlvit/chunk_3.txt

The abstract from the paper is the following:
Combining simple architectures with large-scale pre-training has led to massive improvements in image classification.