Brandon Smock

We probably should have included extra code to create table detection data, but I believe this was left out because we only do table structure recognition in our latest paper....

Hi, The code doesn't currently support this because we didn't design TATR to be used without words. But it makes sense to still want to evaluate the model performance. We...

Hi, Since Rect is in the form (x0, y0, x1, y1), in your case we would have y0 > y1, which means technically the box has negative area. Is this...

If you're training on the original PubTables-1M and FinTabNet.c (FinTabNet.a6) together then one reason you may see lower numbers during evaluation is we changed how we evaluate on PubTables-1M in...

Hi, thanks for bringing this up, it's a great point. If I understand you correctly, this situation can happen when two predicted columns are non-overlapping but stacked vertically, one above...

Great observation, I see the discrepancy you are referring to. The key question is: what do we do when "no-object" is the most probable class prediction? Do we suppress/filter out...

Hi, Is this with one of the pre-trained models and one of the released datasets? If so can you share which model and which input image? If not, is there...

Hi, I think with anything in document information extraction, the first question I would ask is: what would you like the final result to be? That question comes before deciding...

In the current version of the code, you can find the function that takes the model output and processes it into a table representation here: https://github.com/microsoft/table-transformer/blob/3e1dd0c3cad7956c790765b491ec86817e94ce43/src/grits.py#L727

> How can I get the page_tokens if I will use a new Image input Right now the code is written to be used with the PubTables-1M dataset or any...