Skip to main content

Indie game storeFree gamesFun gamesHorror games
Game developmentAssetsComics
SalesBundles
Jobs
TagsGame Engines

GameTranslate

In-game translator at your disposal · By Godnoken

About The OCR Engine

A topic by mfilv created 75 days ago Views: 94 Replies: 5
Viewing posts 1 to 3

Will You Add More Powerful OCR Engine in the future ?

(1 edit)

Another question, will you add paddle ocr in the future ?

Developer (1 edit)

GameTranslate uses RapidOCR, which is essentially PaddleOCR in a different format. They use the same models (Paddle's models). However, there could be differences in their output due to conversion errors. My implementation of RapidOCR is likely not perfect, either.

Unfortunately, the Paddle models only work for CJK & English at the moment. There are some Paddle models for other languages out there in the wild, but some of them seem broken and I just haven't had time to test all of them. The English models do have issues with spacing sometimes too. I'm looking to update the English & Chinese model to the new ones they just released a week ago. Hopefully it will be done soon.

In regard to other OCR engines - Yes, this is likely to happen, although they may not be more powerful. I would likely add another one just to be able to support more languages as Tesseract is subpar when it comes to anything non-documented-like.

Developer

Good news since you're using the Arabic models. The next version coming on Sunday will contain a text recognition model for Arabic. It is certainly better than the Tesseract model, but not quite as impressive as the other RapidOCR models. Hopefully we will see more fine-tuned v5 Paddle models for other languages soon.

Developer

The Arabic model is in the new update;

https://godnoken.itch.io/gametranslate/devlog/957434/update-043-arabic-rtl-suppo...

(1 edit)

I’m wondering what kind of issues you are encountering if you are asking for a more powerful OCR engine ? Frankly, the OCR is near perfect, but I only use it for English translations.

Although I have to admit, Paddle OCR seems very impressive for Kanji based languages.