Docvqa data and its format

[omeryasar]

Hi, Thanks for sharing your valuable work.
I try to finetune dessurt model to my own dataset. Contains receipt like documents. What is the exact supervision that ı need to provide. And Do ı have to give bounding boxes of answers etc... I would be so happy if you share the dataset you used or format of the dataset you used for docqva task.

Thanks in advance

[herobd]

If you just want textual answers/supervision, the MyDataset class should meet your needs: https://github.com/herobd/dessurt#mydataset
It just requires a very simple json annotation format. The only tricky part is picking the task token, which depends on what your doing. "natural_q~" is the one for answering questions like DocVQA

[omeryasar]

If you just want textual answers/supervision, the MyDataset class should meet your needs: https://github.com/herobd/dessurt#mydataset It just requires a very simple json annotation format. The only tricky part is picking the task token, which depends on what your doing. "natural_q~" is the one for answering questions like DocVQA

Thanks for the answer, so I dont have to locate my questions or answers with bounding boxes in related document ?

[herobd]

Correct