Skip to content

Latest commit

 

History

History
77 lines (57 loc) · 1.97 KB

README.md

File metadata and controls

77 lines (57 loc) · 1.97 KB
Raw

Overview

top

Manga Global is a simple Manga Translation System. It generates a translated GIMP format data from an image.

Setup

GIMP

$ sudo apt install gimp

# python-fu
$ sudo apt install python-cairo python-gobject-2
$ wget http://mirror.ip-projects.de/ubuntu/pool/universe/p/pygtk/python-gtk2_2.24.0-6_amd64.deb
$ wget http://ftp.de.debian.org/debian/pool/main/g/gimp/gimp-python_2.10.8-2_amd64.deb
$ sudo dpkg -i python-gtk2_2.24.0-6_amd64.deb
$ sudo dpkg -i gimp-python_2.10.8-2_amd64.deb

Python

$ pip install -r requirements.txt

Google Vision API

We use OCR of Google Vision API. So first, you need to set up Google Vision API Environment. https://cloud.google.com/vision

You need to get a service account key, and set the key path to OCR_KEY_PATH in private_const.py.

DEEPL API

We use DeepL. You need to get DeepL API Key, and set DEEPL_AUTH_KEY in private_const.py.

Font

For kivy GUI System, download mplus-2c-regular.ttf, put under fonts. https://www.wfonts.com/font/m-2c

$ mkdir fonts
$ mv path/to/mplus-2c-regular.ttf fonts

For GIMP Output, install Helsinki. https://www.dafont.com/helsinki.font

How to Use

1. Detect texts by OCR

Put images under some directory (INPUT_DIR), like sample_imgs.
Run the following command, you can get ocr result images and jsons.

$ python ocr.py -i <INPUT_DIR> -o <OCR_OUTPUT_DIR>

ocr_result_sample

2. (Additional Step) Fix OCR Results

Currently, OCR precision is not good enouth. It is better to fix results in gui. you can

  • Remove abundant Box.
  • Fix text contents.
$ python refine.py <OCR_OUTPUT_DIR>/jp_json

gui

3. Translation

Run the following command, you can get gimp format data under OCR_OUTPUT_DIR/gimp.

$ python translate.py -i <OCR_OUTPUT_DIR> --is_translate

gimp

Output Setting

You can control output font {size, type}, {input, output} language in const.py