Pubtabnet's structure token annotations differ from origin

[Arlen-yuzu]

In mini_pubtabnet_examples.jsonl, structure tokens have '[', ']'，like '[', ']', which is different from origin annotations in pubtabnet. Could you share your data processing python scripts about this.

[gjgjh]

This issue is the same as #30 . Unitable has preprocessed the original Pubtabnet dataset, using <td>[]</td> to represent a non-empty cell and <td></td> to represent an empty cell. The processed format is similar to mini_pubtabnet_examples.jsonl. Once familiar with the data format, this preprocessing task should not be difficult to implement.