What made me decide to make this?
I love manga, but can’t read Japanese. And Google Translate isn’t so great with Japanese text localization and doesn’t offer a free solution for OCR+translation. So I decided to build something that’ll help me translate the manga more efficiently into English. Additionally, the technology to detect the speech bubbles could also help official translators translate manga faster. Datasets in this space are rare and the largest I found Manga109 is hard to access and only available for academic research. I want to create a free and publicly available artificial dataset.
You can find the documentation for this code here
Associated ML repositories (Work in progress)