CoolFace
Datasetpublic

bluolightning/manga109s-line-annotations

Manga109-s Text Line Annotations High-precision, line-level bounding box and polygon annotations for the Manga109-s Dataset, supporting both full manga pages and speech bubble crops. Furigana is not labeled and is almost entirely excluded from line labels. Includes 8-point oriented polygons for slanted/rotated text lines. The annotation process is documented in METHODOLOGY.md (WIP). Notice: This dataset contains zero dialogue text and zero images. It requires your own local… See the full description on the dataset page: https://huggingface.co/datasets/bluolightning/manga109s-line-annotations.

sourceHugging Facemitupdated 4d agoView on Hugging Face
0likes26downloads
Dataset Card

Manga109-s Text Line Annotations

High-precision, line-level bounding box and polygon annotations for the Manga109-s Dataset, supporting both full manga pages and speech bubble crops. Furigana is not labeled and is almost entirely excluded from line labels. Includes 8-point oriented polygons for slanted/rotated text lines. The annotation process is documented in METHODOLOGY.md (WIP).

Notice: This dataset contains zero dialogue text and zero images. It requires your own local copy of Manga109-s (or Manga109) to reconstruct dialogue text and process images.

Dataset Statistics

MetricCount
Manga Books87
Manga Pages380
Speech Bubble Crops5,988
Text Lines14,182
Oriented Polygons392
Total Characters74,506
Average Lines / Bubble2.37
Average Characters / Line5.25

Quickstart

1. Prerequisites & Installation

Obtain an official copy of Manga109-s from Hugging Face (`hal-utokyo/Manga109-s`) (or the official Manga109 website; the 2026 version is recommended).

Ensure you have Python 3.8+. You can install the companion package or run directly with uv:

bash
# Option A: Install as an editable package (provides the `manga109-lines` CLI)
pip install -e .
manga109-lines --help

# Option B: Run directly with uv (recommended, no manual install needed)
uv run manga109-lines --help
# or
uv run python build_dataset.py --help
Note: In all subsequent examples, manga109-lines and uv run python build_dataset.py can be used interchangeably.

2. Verify Alignment with Local Manga109-s

Check that your local Manga109-s files match the annotation geometry:

bash
uv run python build_dataset.py verify --manga109-dir ./manga109s-v2026

3. Reconstruct Full Annotations with Dialogue Text

Fills in the official dialogue text from your local Manga109-s XMLs or CSV into a complete verified_data.json:

bash
uv run python build_dataset.py reconstruct \
  --manga109-dir ./manga109s-v2026 \
  --output verified_data_reconstructed.json

Exporting to Machine Learning Formats

1. YOLO Format (Ultralytics YOLOv8 / YOLO11)

Mode A: Full Page Line Detection

Detects text lines across entire manga pages (images/{book}/{page}.jpg):

bash
uv run python build_dataset.py export-yolo \
  --manga109-dir ./manga109s-v2026 \
  --target page \
  --task detect \
  --output-dir yolo_line_pages
Mode B: Bubble Crop Line Detection

Detects individual text lines within cropped speech bubbles (crops/{id}.png):

bash
uv run python build_dataset.py export-yolo \
  --manga109-dir ./manga109s-v2026 \
  --target crop \
  --task detect \
  --output-dir yolo_line_crops

Options:

  • —--task segment: Exports normalized 8-point polygon segmentations for oriented lines.
  • —--include-images: Automatically copies or symlinks images into images/train and images/val.

2. COCO Instances JSON

Exports standard COCO instances JSON with text_line (category 1) and text_block (category 2):

bash
# Page-level COCO
uv run python build_dataset.py export-coco \
  --manga109-dir ./manga109s-v2026 \
  --target page \
  --output manga109s_lines_coco_page.json

# Crop-level COCO
uv run python build_dataset.py export-coco \
  --manga109-dir ./manga109s-v2026 \
  --target crop \
  --output manga109s_lines_coco_crop.json

3. Enhanced Manga109 XML Files

Inserts <line index="..." xmin="..." ymin="..." xmax="..." ymax="..."> tags directly into the official Manga109 XML files:

bash
uv run python build_dataset.py export-xml \
  --manga109-dir ./manga109s-v2026 \
  --output-dir manga109s_xml_with_lines

Sample output element:

xml
<text id="00000d6f" xmin="192" ymin="957" xmax="263" ymax="1036">セリフ１\nセリフ２\nセリフ３
  <line index="0" xmin="236" ymin="957" xmax="257" ymax="1014">セリフ１</line>
  <line index="1" xmin="213" ymin="957" xmax="235" ymax="1036">セリフ２</line>
  <line index="2" xmin="192" ymin="957" xmax="211" ymax="1036">セリフ３</line>
</text>

4. Line OCR Dataset (JSONL)

Exports a line-level OCR mapping file for text recognition training:

bash
uv run python build_dataset.py export-ocr \
  --manga109-dir ./manga109s-v2026 \
  --output manga109s_crops_ocr.jsonl

Data Schema (line_annotations.json)

json
{
  "images/ARMS/065.jpg": {
    "book": "ARMS",
    "page_index": 65,
    "width": 1654,
    "height": 1170,
    "texts": [
      {
        "id": "00000d6f",
        "xmin": 192,
        "ymin": 957,
        "xmax": 263,
        "ymax": 1036,
        "line_lengths": [3, 4],
        "lines": [
          {
            "line_index": 0,
            "xmin": 236,
            "ymin": 957,
            "xmax": 257,
            "ymax": 1014,
            "char_count": 3
          },
          {
            "line_index": 1,
            "xmin": 213,
            "ymin": 957,
            "xmax": 235,
            "ymax": 1036,
            "char_count": 4,
            "polygon": [213, 960, 230, 1036, 235, 1033, 218, 957]
          }
        ]
      }
    ]
  }
}

Key Fields

  • —line_lengths: Array of character slice lengths for each line in reading order.
  • —char_count: Number of characters corresponding to this line.
  • —polygon: (Optional) 8-point coordinate array [x1, y1, x2, y2, x3, y3, x4, y4] specifying the oriented bounding box for tilted or slanted lines (present on 392 lines).

License & Citation

Annotation & Code License

The line annotations and accompanying tooling (build_dataset.py) are released under the MIT License.

Note: This license applies solely to the line annotation geometry files and associated utility code. The underlying Manga109-s dataset and artwork remain subject to the Manga109 Terms of Use.

Manga109 Terms of Use Notice

This release strictly abides by the Manga109-s Terms of Use. To use this dataset with original text or imagery, you must obtain a legitimate copy of Manga109-s from:

Citation

If you use these line annotations or conversion tools in your research, please cite this repository:

bibtex
@misc{bluolightning2026manga109slines,
  author       = {Nav (bluolightning)},
  title        = {{Manga109-s Text Line Annotations}},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/bluolightning/manga109s-line-annotations}}
}

Please also cite the underlying Manga109 / Manga109-s dataset and annotation papers:

@inproceedings{baek2026mangav26,
  title     = {{Manga109-v2026: Revisiting Manga109 Annotations for Modern Manga Understanding}},
  author    = {Baek, Jeonghun and Miyai, Atsuyuki and Onohara, Shota and Ikuta, Hikaru and Aizawa, Kiyoharu},
  booktitle = {Culture × AI Workshop at ICML 2026},
  year      = {2026}
}

@article{multimedia_aizawa_2020,
  author  = {Aizawa, Kiyoharu and Fujimoto, Azuma and Otsubo, Atsushi and Ogawa, Toru and Matsui, Yusuke and Tsubota, Koki and Ikuta, Hikaru},
  title   = {Building a Manga Dataset ``{Manga109}'' with Annotations for Multimedia Applications},
  journal = {IEEE MultiMedia},
  volume  = {27},
  number  = {2},
  pages   = {8--18},
  doi     = {10.1109/mmul.2020.2987895},
  year    = {2020}
}

@article{mtap_matsui_2017,
  author  = {Matsui, Yusuke and Ito, Kota and Aramaki, Yuji and Fujimoto, Azuma and Ogawa, Toru and Yamasaki, Toshihiko and Aizawa, Kiyoharu},
  title   = {Sketch-based Manga Retrieval using {Manga109} Dataset},
  journal = {Multimedia Tools and Applications},
  volume  = {76},
  number  = {20},
  pages   = {21811--21838},
  doi     = {10.1007/s11042-016-4020-z},
  year    = {2017}
}