nangleraud commited on
Commit
1d0ca7d
Β·
verified Β·
1 Parent(s): 95623f7

Update README.md

Browse files

Correct approximate number of synthetic Greek and Latin pages generated in the dataset

Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -27,7 +27,7 @@ The project operates along three major axes:
27
 
28
  ## Technical Approach
29
 
30
- A core challenge is the diversity of historical typesettings and complex page layouts found in critical editions (main text, apparatus criticus, footnotes, canonical references, multi-column layouts). To address this without prohibitively expensive manual annotation, we generate synthetic training data rendered from openly available texts β€” currently approximately 1,400,000 Greek pages and 50,000 Latin pages, drawn from 4,582 works, with over 5,000 style combinations.
31
 
32
  The pipeline is: **OCR β†’ TEI XML β†’ Quality Assurance β†’ Distribution**.
33
 
 
27
 
28
  ## Technical Approach
29
 
30
+ A core challenge is the diversity of historical typesettings and complex page layouts found in critical editions (main text, apparatus criticus, footnotes, canonical references, multi-column layouts). To address this without prohibitively expensive manual annotation, we generate synthetic training data rendered from openly available texts β€” currently approximately 174,694 Greek pages and 10,000 Latin pages, drawn from 4,582 works, with over 5,000 style combinations.
31
 
32
  The pipeline is: **OCR β†’ TEI XML β†’ Quality Assurance β†’ Distribution**.
33