anonimkaq4/manga-page-element-segmentation

Manga Page Element Segmentation

YOLO11m instance-segmentation model for manga pages: characters, panel frames, and speech bubbles.

What the model recognizes

The model predicts an instance mask, bounding box, class label, and confidence score for every detected object. It has three classes:

Output label	Meaning	Typical use
character	A visible person or character.	Find people for cropping, composition analysis, or character-aware editing.
frame	A manga panel/frame. The mask follows the visible panel area or border.	Split a page into panels before reading-order or per-panel analysis.
speech_bubble	A dialogue/speech balloon.	Send dialogue regions to OCR or exclude them from artwork processing.

Use the literal labels above when reading result.names from Ultralytics. Class IDs are 0: character, 1: frame, and 2: speech_bubble.

snapshots/yolo11m_1024_synthetic_mangaseg/best.pt is the recommended checkpoint.

Checkpoints

Checkpoint	Base model	Training snapshot	Mask mAP50	Mask mAP50-95
snapshots/yolo11m_1024_synthetic_mangaseg/best.pt	yolo11m-seg.pt	epoch 33	0.8689	0.6620
snapshots/yolo11m_1024_b2/manga-yolo11m-seg-validation-best.pt	yolo11m-seg.pt	epoch 15	0.8742	0.6078
snapshots/yolo11s_1024_b8/manga-yolo11s-seg-validation-best.pt	yolo11s-seg.pt	epoch 18	0.8649	0.5895

The recommended checkpoint was evaluated at 1024 px on 784 held-out MangaSeg/Manga109-s pages.

Class	Mask mAP50	Mask mAP50-95
character	0.7731	0.5829
frame	0.8427	0.5164
speech_bubble	0.9908	0.8867

Quick start

pip install ultralytics huggingface_hub

from huggingface_hub import hf_hub_download
from ultralytics import YOLO

checkpoint = hf_hub_download(
    repo_id="anonimkaq4/manga-page-element-segmentation",
    filename="snapshots/yolo11m_1024_synthetic_mangaseg/best.pt",
)

model = YOLO(checkpoint)
results = model("manga_page.png", imgsz=1024, conf=0.25)

for result in results:
    print(result.names)  # {0: 'character', 1: 'frame', 2: 'speech_bubble'}
    result.save(filename="prediction.png")

Older YOLO11m and YOLO11s snapshots remain available for comparison.

Usage and license

This model was trained on MangaSeg annotations derived from Manga109-s. This repository does not include or grant rights to the original Manga109-s images or MangaSeg annotations. Review and comply with the MangaSeg, Manga109-s, and Ultralytics licenses before using the underlying datasets, publishing dataset-derived images/results, redistributing this checkpoint, or using it commercially.

If using MangaSeg images or derived visual results, include the credit: "Copyrighted by Minshan Xie".

Citation

This model was trained using MangaSeg annotations. If you use the model, its training data, or derived results, cite the MangaSeg paper:

@inproceedings{xie2025advancing,
  title={Advancing Manga Analysis: Comprehensive Segmentation Annotations for the Manga109 Dataset},
  author={Xie, Minshan and Lin, Jian and Liu, Hanyuan and Li, Chengze and Wong, Tien-Tsin},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2025}
}

Also cite Manga109-s if your work uses or reports results derived from the original Manga109-s pages or layout annotations:

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