TextRecognitionDataGenerator hindi text render incorrect

When I try to use this repo to generate hindi text imgs, it works well. But actually all characters are incorrect because PIL.trutype could only render a simple character-to-glyph but hindi text is complex.

here is the same problem: https://github.com/python-pillow/Pillow/issues/3191

Jun 27 '19 04:06 KosukeHao

So it's pretty much like arabic where ligatures were not used properly.

I'll take a look at that Pillow issue, they seem to have linked to a few library that address the issue.

Thank you for reporting this.

Jun 27 '19 10:06 Belval

@Belval @KosukeHao how can I generate Hindi images. It is mentioned that language that we can use should be in French, English, Spanish, German or Chinese

Mar 04 '20 16:03 dwivediagam

There is no solution as of today. Unfortunately, the ligatures are still not supported and needed dependencies are unknown.

I would be very interested if you can make a PR that has a working sample. I would love to work on it myself, but I do not know Hindi and cannot see the difference between good and bad samples. If you can provide clear examples of what TRDG should output given a specific input I could give it another try.

Mar 04 '20 18:03 Belval

Installing libraqm as suggested here should help in most cases: https://stackoverflow.com/questions/39630916/

Edit:
Sometimes installing libraqm causes the following error:
OSError: invalid face handle Sometimes it just works well. Not sure what could be the cause.

Nov 04 '20 08:11 GokulNC

Is this issue solved? I would like to use the library to generate in Indian Punjabi (Gurmukhi) language which is similar to Hindi (Devnagari). Please let me know if you need some help in Punjab language.