surya
surya copied to clipboard
OCR of vertical Japanese text
It seems that Surya OCR in the default settings cannot read vertical Japanese text correctly. Is there any settings that I can use to improve the results?
It seems that Surya OCR in the default settings cannot read vertical Japanese text correctly. Is there any settings that I can use to improve the results?
It's terrible.
Also, this OCR works terribly on manga. Too many errors, and unlike Paddle OCR, there's no way to get the "weight" of the recognized text. It's better with full-color images, but if the speech bubble has a transparent background, I recommend choosing something else. Without a software API, it’s completely unclear how to integrate this anywhere.
Someday the model will be trained ...