chunked/nltk_chunking/model_doc_visual_bert/chunk_12.txt

For the VCR task, the authors use a fine-tuned detector for generating visual embeddings, for all the checkpoints.