chunked/nltk_chunking/model_doc_owlvit/chunk_4.txt

For object detection, pre-training and scaling approaches are less well established, especially in the long-tailed and open-vocabulary setting, where training data is relatively scarce.