chunked/nltk_chunking/model_doc_visual_bert/chunk_19.txt

The authors use the features generated after passing these regions through a pre-trained
CNN like ResNet as visual embeddings.