heresta / ocr17plus Goto Github PK

Thank you for sharing this corpus.
Creating GT is not an easy job. I took a random look at the page files from the pageXmlTranskribusCorrected folders.

I noticed the following problems:

1. the entire text of a line was encoded at the Word level, as a single Word.
   Solution: Convert Word ind line
2. often the drop-capital are annotated as Graphic
3. many separators can be seen as so called fake separators and should be corrected
4. a wish, Transkribus does not create valid page instances, of course such annotations as:
   <TranskribusMetadata docId="188203" .../> can be commented out.
   but:

• open type="" attributes
• open id="" Attributes should be corrected to.
• • the Alto format files contain very deeply structured data, unfortunately when converting to Page-XML format this information was not included.

I will be very welcome to help you to improve the data within my possibilities.
Thanks again for everything

If I am not mistaken, you are checking for full path equivalency (

I think this should be checked but I have some doubts here.

For the file Baron1686_Enlevemens_bpt6k73945k/alto4eScriptorium/44_07767_default.xml, the image is missing.