You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It would be beneficial to make OCR models optional during installation, with EasyOCR remaining as the default option. In our case, we use TesseractOCR but are required to install EasyOCR since it's currently mandatory, even though we don't use it.
Here's a proposed installation approach:
All OCR models: pip install docling[all]
EasyOCR only (default installation): pip install docling[easyocr]
Specific OCR models: pip install docling[tesseract]
Base installation (no OCR models): pip install docling
Alternatives
Install Docling as is - This installs EasyOCR and its dependencies even when they're not needed.
Install Docling without dependencies - This requires significant maintenance effort on our end to ensure version compatibility.
The text was updated successfully, but these errors were encountered:
@jaluma We have been considering different options, including no OCR as default, but we decided against that. To have a seamless out-of-the-box experience, EasyOCR is installed always as a dependency but no OCR models are downloaded unless you actually use EasyOCR the first time. This keeps the overhead minimal, since most of EasyOCRs dependencies are anyway shared with other packages we need.
Requested Feature
It would be beneficial to make OCR models optional during installation, with EasyOCR remaining as the default option. In our case, we use TesseractOCR but are required to install EasyOCR since it's currently mandatory, even though we don't use it.
Here's a proposed installation approach:
All OCR models:
pip install docling[all]
EasyOCR only (default installation):
pip install docling[easyocr]
Specific OCR models:
pip install docling[tesseract]
Base installation (no OCR models):
pip install docling
Alternatives
The text was updated successfully, but these errors were encountered: