DocIntel table data doesn't work correctly

[Sergey Popov]

Hello,

I'm trying to train a DocIntel model to recognize table data of Purchase Orders.

Using the DemoHub's sample purchase orders which I printed and scanned.

I'm able to train them for static data, but not for tables. 

I always see some weird results like in the attachment. 

It looks like the engine is not aware of table lines, that data should be aligned in line. 

Some data is taken from a wrong line, and some data is taken not from the table at all!

Also, I wasn't able to teach it to recognize more than one line...

Is line recognition function something that we should not rely upon yet?

[Loic1]

Hi Sergey,

Can you provide more details about the demo narrative you are using?

I don't see an Item Code on the table data for the example you are sharing, so it makes sense that the value is being predicted from other areas of the document.

[Sergey Popov]

Hi Loic1,

The narrative is named "Accelerate citizen automation to unleash productivity and innovation with a low code experience powered by OOTB template: (Tokyo)". 

But this out of narrative's bounds, there's no table data used.
I was just trying to train the model to recognize table data and failed, because the system is likely not aware of table structure.

[Loic1]

As mentioned, the "Item Code" field does not exist in your table, the other fields are processed properly and the table structure is identified correctly.

[Sergey Popov]

1. Yes, the Item Code is missing in the table, but it was marked as "missing in this document" on training.

2. The system haven't recognized the other lines of the table. Only the first one.

3. The same issue happens on different documents. It's always recognizes a single line of the table, and fetches data from different document lines to that single line.