Package: wnpp
Severity: wishlist

* Package name    : tesseract-ocr-deu-best
  Version         : upstream traineddata snapshot
  Upstream Author : Tesseract OCR project
* URL             : https://github.com/tesseract-ocr/tessdata_best
* License         : Apache-2.0
  Programming Lang: N/A (trained OCR language data)
  Description     : High-accuracy German OCR language model for Tesseract

The Tesseract OCR project provides two major LSTM language
model variants: tessdata_fast and tessdata_best.

Debian 13 currently ships tessdata_fast through packages
such as tesseract-ocr-deu and tesseract-ocr-eng.

The upstream tessdata_best models offer higher potential
OCR accuracy at the expense of processing speed.

I would like to request official Debian packaging of the
tessdata_best language models.

Suggested improvements:

- Provide packages such as tesseract-ocr-deu-best and
  tesseract-ocr-eng-best.
- Allow Fast and Best models to coexist without conflicts.
- Manage installation, updates and removal through APT.
- Preserve the existing Fast models as the default.
- Document how applications can select the Best models.
- Consider packaging other languages consistently.

This would be especially beneficial for document management
systems such as Paperless-ngx, where OCR accuracy can be
more important than processing speed.

Currently, users must install Best models manually or
replace files owned by Debian packages.

Official packages would improve maintainability and
upgrade safety.

Related Paperless-ngx feature request:
https://github.com/paperless-ngx/paperless-ngx/issues/14789

Upstream license:
https://github.com/tesseract-ocr/tessdata_best/blob/main/LICENSE

I am requesting this package as a user and am not
volunteering to maintain it.

Reply via email to