Availability of pretrained model for English (openocr_svtrv2_en)?

[pvd0anh]

Hello authors,

Thank you for your great work on OpenOCR and SVTRv2.

I noticed that openocr_svtrv2_ch is available for Chinese.

Is there a specific pretrained checkpoint/config released for English?

I intend to fine-tune on a Latin-based dataset (Vietnamese), so the English weights would be much better than the Chinese ones.

Thanks!

[alephpi]

Since this model recognizes English as well, I think you can just distill the Chinese tokens away from the embedding and lm_head layers, then extend the tokenizer and these two layers with your Vietnamese vocab dictionary.

[pvd0anh]

@alephpi Thank you for the suggestion.
I understand that I can modify the embeddings and lm_head and extend the tokenizer with my Vietnamese vocabulary.
However, I would like to confirm whether there is any alternative version or pre-adapted model available, aside from manually customizing it myself.