mangaOCR

See:

1. https://github.com/kha-white/manga_ocr
2. xulihang/ImageTrans-docs#140

Note: If your device is powerful enough, it is recommended to use manga-image-translator.

Installation

The ONNX version of mangaOCR is included since ImageTrans v4.2.0, so you don't need to do any installation. You can find the mangaocr folder under ImageTrans's root. If there is not such a folder, you can create one and put decoder.onnx, encoder.onnx and vocab.txt in it.

If you need to use Python to have a better performance, you can read the following:

1. Remove the mangaocr folder under ImageTrans's root.
2. Install Python3 if it is not installed.
3. Install required packages: pip3 install bottle==0.12.19 manga_ocr==0.1.4 Paste. (Use terminal to run the commands.)
4. Download the model and unzip it into a folder.
5. Run the server: python3 server_manga_ocr.py.
6. Unzip the plugin files in ImageTrans's plugins folder and restart ImageTrans. (optional for v1.9.0+)

For convenience, you can also use the packed versions:

• Windows
  1. Download and unzip manga-ocr.
  2. Download the model and unzip it to manga-ocr's folder in the previous step. (if you do not download the model file, the program will download the model from the Internet.)
  3. Run run.bat and wait for the server to get ready. If you need to enable GPU, please follow this guide.
• macOS: https://github.com/xulihang/ImageTrans_plugins/releases/download/plugins/mangaOCR.dmg

Usage Note

The mangaOCR works great for speech bubbles like the following one:

But it may not work well for the following long text line image by default:

For such a case, you can select the long text mode of the plugin which will crop the long text line images into segments for the OCR engine to extract the text. It should better be used in combination with a scene text detector which detects the text lines accurately like this one: https://github.com/xulihang/ImageTrans_plugins/tree/master/mangaTranslatorOCR

Text Detection Combination

mangaOCR does not detect text. It can be used together with other text detection methods like the following ones:

1. Speech bubble detection models: https://github.com/xulihang/balloon-dataset
2. manga image translator (detect text lines): https://github.com/xulihang/ImageTrans_plugins/tree/master/mangaTranslatorOCR

Port Settings

You may need to update the port if you also need to run other local OCR engines:

https://github.com/xulihang/ImageTrans_plugins/tree/master#notes-using-local-servers