#370 - Feature request - OCR - FoundKeyGang/FoundKey

jeder commented

2023-03-29 18:44:32 +00:00

Contributor

I just got reminded of that, and thought that it would be a nice to have ~~and I know that it won't be added any time soon but whatever lmao~~

Anyway, OCR for alt-text yes
I am lazy and can't be arsed to write a wall of text yes

I just got reminded of that, and thought that it would be a nice to have ~~and I know that it won't be added any time soon but whatever lmao~~ Anyway, OCR for alt-text yes I am lazy and can't be arsed to write a wall of text yes

Johann150 added the

feature

label 2023-03-30 15:43:58 +00:00

Johann150 commented

2023-03-30 15:47:35 +00:00

Owner

Mastodon seems to use client side Tesseract, but that apparently requires adding a 10MiB GZip-ed blob to your web app. Not sure I'm a fan of that.

Are there some API providers maybe?

Mastodon [seems to use](https://github.com/mastodon/mastodon/pull/11566) client side Tesseract, but that apparently requires adding [a 10MiB GZip-ed blob](https://github.com/mastodon/mastodon/blob/main/public/ocr/lang-data/eng.traineddata.gz) to your web app. Not sure I'm a fan of that. Are there some API providers maybe?

jeder commented

2023-03-31 22:38:59 +00:00

Author

Contributor

all of the stuff i've found so far just seems scummy 🥴

thatonecalculator commented

2023-04-10 01:33:42 +00:00

Feel free to use/reference https://codeberg.org/calckey/calckey/commits/branch/main/search?q=ocr for our implementation. It uses Tesseract, but on the backend.

mia commented

2023-05-31 17:44:19 +00:00

Contributor

If you’re gonna run OCR on the backend, might also consider EasyOCR on especially beefy systems. It’s much better in my experience, except it doesn’t support vertical Chinese/Japanese text.

Sadly there’s no library, only a Python module.

If you’re gonna run OCR on the backend, might also consider [EasyOCR](https://github.com/JaidedAI/EasyOCR) on especially beefy systems. It’s much better in my experience, except it doesn’t support vertical Chinese/Japanese text. Sadly there’s no library, only a Python module.

👀 1

Johann150 closed this issue

2023-08-04 16:27:12 +00:00

Feature request - OCR #370