ndl-lab/reading_order

NDLOCR(version2)用読み順推定モジュール

reading_order(NDLOCR(ver.2.1)用読み順推定モジュール)

NDLOCR(ver.2.1)用の読み上げ順序推定モジュールのリポジトリです。

本プログラムは、令和4年度NDLOCR追加開発事業の成果物であるver.2.0に対して、国立国会図書館が改善作業を行ったプログラムです。

事業の詳細については、令和4年度NDLOCR追加開発事業及び同事業成果に対する改善作業をご覧ください。

本プログラムは、国立国会図書館がCC BY 4.0ライセンスで公開するものです。詳細については LICENSEをご覧ください。

Installation

To install the program, first navigate to the cloned directory and install the required packages. It is recommended that you create an appropriate environment using venv or conda before proceeding with the installation process.

$ cd reading_order
$ pip install .

Reading Order Detection from XML Files

To detect reading order from XML files, organize the input data in the following folder structure, where pid is a unique identifier for each document:

|- XXXXXX (pid)
|  |- xml
|  |  |- XXXXXX.xml
|  |- img (optional)
|     |- *.jpg
|- ...

After organizing the input data, run the following command:

$ python tools/eval.py --skip .sorted.xml XML/DIR

Replace XML/DIR with either the unsorted XML file or the folder containing unsorted XML files. If a folder is specified, the program will search for XML files recursively. The above command generates a .sorted.xml file in the same directory as the original .xml file.

|- XXXXXX (pid)
|  |- xml
|  |  |- XXXXXX.xml
|  |  |- XXXXXX.sorted.xml
|  |- img (optional)
|     |- *.jpg
|- ...

Visualize the Results

You can visualize the results by running the following command:

$ python tools/visualize.py XML/DIR --only .sorted.xml

This generates a .order.jpg file in the same directory as the original .jpg file.

|- XXXXXX (pid)
|  |- xml
|  |  |- XXXXXX.xml
|  |  |- XXXXXX.sorted.xml
|  |- img (optional)
|     |- *.jpg
|     |- *.order.jpg
|- ...

Alternatively, you can output all the .order.jpg files in one place DIR by using the --output option as follows:

$ python tools/visualize.py XML/DIR --only .sorted.xml --output DIR