Note: If you want to use a Dedicated AI Engine make sure you have purchased the feature before you start the training process.
Below, you can find some additional instructions and practical examples that may come in handy during your annotation process and help you increase the extraction accuracy of your Dedicated AI Engine.
1. Annotate only the values, not the labels.
2. Move the bounding box a little.
If the bounding box crosses right through a word, adjust it. This will help you eliminate partial word extractions.
3. Annotate the totals where written as the sum of all line items.
Do not annotate these within the line items table, even if the value is the same.
4. Annotate the values on the locations close to the related data.
For example, if a vendor name is located in the upper part of the document together with his address and ID, annotate it there rather than taking his name from the footer of the document.
5. Annotate the values written with usual font.
If you have the the same value written in the logo, in the footer and in the body of a document, choose the one with more usual font and size.
6. If a field is not a line item field, do not annotate it as a line item field.
Header fields should be annotated as header fields and line items as line items, unless Rossum instructed you differently. Typically, this is often a case of the amounts being the same as a line item amounts if there is only one line item on the invoice. But even though the amount may be the same, annotate it within the table footer, not within the line item.
7. Annotate taxes to a tax table if located in a tax table and to the header fields if not located in a tax table on the document.
There are two different ways to annotate tax details: using a tax details table and using relevant header fields (Total Tax, Tax Code, etc.). The tax details table is there for cases when taxes are written in a horizontally placed table, or if there are multiple tax rates (information) you need to distinguish. Example of a tax table:
Example of tax placed in the invoice totals – this is not a tax table, it is a document total table:
Values located in the tax table in the document need to be annotated in the tax table, not header fields. Values outside the tax table need to be annotated in the header fields, if there is only one type of tax. If there are multiple taxes (and related data), such data cannot be annotated to one header field, so they need to be annotated in a tax table as well.
To move around the Rossum app quickly, consider using keyboard shortcuts.