chunked/nltk_chunking/model_doc_owlvit/chunk_7.txt

Our analysis of the scaling properties of this setup shows that increasing image-level pre-training and model size yield consistent improvements on the downstream detection task.