Hacker News Reader

If you don't mind disclosing, what was your particular use-case (the labeled dataset you trained on)?

It was for digitizing 19th century books written in a font and language not supported in a vanilla Tesseract.