The models trained in our baseline are listed below. All models were trained 1 epoch under their respective backbones using the pretrained models provided by Transformers, LayoutLM and detectron2.
The models trained on DocBank are available in the format used by Pytorch.
| name | backbone | url | size | |
|---|---|---|---|---|
| 0 | BERT | BERT-base | Azure | 387MB |
| 1 | BERT | BERT-large | Azure | 1.2GB |
| 2 | RoBERTa | RoBERTa-base | Azure | 441MB |
| 3 | RoBERTa | RoBERTa-large | Azure | 1.2GB |
| 4 | LayoutLM | LayoutLM-base | Azure | 398MB |
| 5 | LayoutLM | LayoutLM-large | Azure | 1.2GB |
| 6 | X101 | ResNeXt-101 | Azure | 747MB |