chunked/nltk_chunking/model_doc_layoutlmv2/chunk_24.txt

Important to note is that the visual backbone expects BGR channels instead of RGB, as all models
  in Detectron2 are pre-trained using the BGR format.