Support hebrew characters

[uriva]

🚀 The feature

Support hebrew characters

Motivation, pitch

Increase user base

Alternatives

No response

Additional context

No response

[felixT2K]

Hi @uriva 👋🏼,

Thanks for the feature request.
Feel free to open a PR and add the vocab in:
https://github.com/mindee/doctr/blob/main/doctr/datasets/vocabs.py
and update the corresponding documentation in:
https://github.com/mindee/doctr/blob/main/docs/source/modules/datasets.rst#supported-vocabs

Afterwards you can train your own model see:
https://mindee.github.io/doctr/using_doctr/custom_models_training.html

[uriva]

#1355

How many examples are required to train a new language?

[felixT2K]

Hi @uriva 👋🏼,

This question is not 100 percent clear to answer.
Since Hebrew is based on the English corpus, you can fine tune our already trained models and do not have to train from scratch.

A rule of thumb is that it is highly recommended to validate on real data (even if it might not be that much).
For training you can also try to generate synthtic data.
For example with: https://github.com/clovaai/synthtiger
Or if possible label real data with AWS Textract or Azure Document AI.

~50K should be a good starting value for the beginning.

The current models are trained from scratch (mindee internal dataset / french vocab ~11M word crop images)