Fyn Update: 2020-03
Non blocking pipeline and improved performance
We have brought a number of changes to Fyn's pipeline to improve the usability and quality of our outputs.
Non blocking pipeline
Fyn integrates many automated document processing steps, among which Document classification and Language Classification. Until now, Fyn would only deal with documents for which the type and language were part of its supported capabilities, meaning that if a document of unsupported type or language was sent, Fyn wouldn't proceed to UBL field detection and extraction.
This behavior was changed so that Fyn will try to extract data even from types of documents or documents in languages that aren't yet supposed to be supported. Fyn can therefore recover from the rare events of wrong document type or language classification.
We've also worked hard on improving the performance of Fyn. To do so, we have again improved the document preprocessing. These preprocessing steps, coupled with significant work on improving our training datasets, largely improve the performance of Fyn.
The evolution of Fyn's performance on all the fields for French invoices:
The evolution of Fyn's performance on the 4 most important fields for French invoices: