UiPath Community 2022.10 Stable Release - Document Understanding
This topic goes in-depth about the improvements in Document Understanding. To read about other products, please navigate to the main topic here.
We’re happy to share with you our latest work - all the bits & pieces coming together in our 22.10 release
Migration to .net5 cross-platform
With this release, we’ve finished our efforts of migrating the packages contributing to the Document Understanding functionality to .net5 cross-platform, enabling their usage on linux robots . We’ve migrated everything except for Omnipage, for which the linux support is currently missing and will come in a future release.
Besides the above, we’ve made several improvements to our existing capabilities
With the newest release of the Digitize Activity, one will have the UiPath Document OCR preselected as default OCR engine - allowing user to just adjust the engine if needed & easily get up & running using our own.
We have also worked on exposing the CJK OCR as service - having it now available for usage both in Studio as well as in the Data Manager.
In order to be able to gather more usage data when it comes to users performing Validation Operations in Action Center, we now report the following when working with the Create Validation and Classification Actions Activities:
Besides the above, we’ve improved the Validation Station, reporting the confidence score for each table entry, and the digitization algorithm by fixing reported bugs
Digitization & Extraction algorithm improvements
We also focused on enhancing the digitization algorithm, which means that the digitization of native PDFs may be faster & more accurate than before (applied when the “Apply OCR on PDF” flag is set to “auto” or “false”). Shall you find the above is not happening as we would hope, please do bring it to our attention and report eventual issues
Besides that, we introduced “hybrid OCR” approach, which would enhance our current “auto” option for digitizing documents, by processing native PDFs the 2 steps when a native PDF is identified:
- Extract the native text
- Are there images identified? => Cool, OCR them & extract the text from them as well.
In this way, one benefits from the best results for native PDFs, by natively extracting the printed text & OCRing the images.
Furthermore, we have refactored the Extraction Result object, to enable more flexible and user-friendly usage of it within the workflow by adding a new, simplified way to represent tables: they are now stored separately from other fields, in a flatter structure and with methods that help you access the data. Separately, we have also added new methods that make it easier to consume and modify data in the extraction result.
And to support the work with multi-value fields, we enhanced the ML Extractor & the ML Extractor Trainer, so that these (multi-value) fields sent from the Document Manager can be consumed and used in the activities.
And if the above was not enough, we are happy to report that with this release all our activities are .net6 compatible
Please tell us what you think
Which features excite you the most and you are going to try right away? We want to hear what you think! Please use the button below