Spanish text recognition finetune

[neo133]

Hi,
Kudos to the amazing work.
I wanted to train spanish text for recognition but i only found that it downloads latin infer model when i put lang=es in inference API.
Is there a training model which I can finetune, If not then how can i train spanish characters on my dataset? Should I just change the dict to latin and thats it? or is threre any other method?

[liuhongen1234567]

Hello, the PP-OCRv5 Latin model is actually not a Latin language model, but a Latin language family model. Its dictionary includes the Spanish language dictionary. Therefore, you can directly use the training and dictionary of the Latin model. The training weights for the Latin model can be downloaded from here https://www.paddleocr.ai/latest/en/version3.x/module_usage/text_recognition.html#2-list-of-supported-models.

[neo133]

thank you for the reply.
So as i understand it is only supported in ppocrv5.. older versions of ppocr like v3 v4 isnt there in latin?

[liuhongen1234567]

V4 does not have any multilingual models, but the multilingual model in V3 includes a Latin model named latin_PP-OCRv3_mobile_rec. If you are using V3, you can check whether its dictionary https://github.com/PaddlePaddle/PaddleOCR/blob/main/ppocr/utils/dict/latin_dict.txt contains the complete set of Spanish characters. However, based on user feedback, the multilingual accuracy of V3 is far inferior to V5.

[neo133]

Thank you so much!!
Also I wanted to ask if Vietnamese characters are supported?