Coder109/C3B

English | 简体中文

---

C³B: Comics Cross-Cultural Benchmark

Culture In a Frame: C³B as a Comic-Based Benchmark for Multimodal Cultural Awareness

ICLR 2026

---

About C³B

C³B (Comics Cross-Cultural Benchmark) is a multicultural, multitask, and multilingual benchmark for evaluating cultural awareness capabilities of Multimodal Large Language Models (MLLMs).

• Progressive task difficulty: From basic visual recognition, to higher-level cultural conflict understanding, to cultural content generation.
• Comic-based: Built on manga/comic images; comics better reflect cross-cultural narratives and conflict scenarios than single-culture real-world images.
• Scale: Over 2,000 images and 18,000+ QA pairs, covering three tasks of increasing difficulty.

Task overview

Type	Description
Basic visual recognition	Recognize and understand culture-related visual elements in comic panels
Cultural conflict understanding	Understand cross-cultural conflicts, cultural symbols, and narrative intent
Cultural content generation	Generate culture-related descriptions or stories from given settings or images

Authors: Yuchen Song, Andong Chen, Wenxin Zhu, Kehai Chen, Xuefeng Bai, Muyun Yang, Tiejun Zhao

---

Data Format

This repository contains C³B annotations and task data in the following structure:

task3/ — Multilingual translation/alignment

• Files: task3.ja-en.json, task3.ja-de.json, task3.ja-es.json, task3.ja-ru.json, task3.ja-th.json, etc.
• Fields: manga_name, page_index, src (source text), tgt (target text).
• Each item is a sentence or phrase pair for translation or alignment between languages.

task12/ — Cultural content generation (Task 1 and Task 2)

• task12 contains the data for Task 1 and Task 2 of the benchmark.
• task12.json: structured data for generation-style evaluation.

task12.json fields (each item in the JSON array):

Field	Type	Description
file_name	string	Image filename (e.g. 0.png)
Q1	object	First question: multiple-choice options as keys A–E (e.g. region/culture names)
A1	string	Answer to Q1 (e.g. one or more option letters such as "B C E")
Q2	object	Second question: options A–E (e.g. lists of scene elements)
A2	string	Answer to Q2 (e.g. one option letter such as "D")
A3	string	Third answer (e.g. "Yes" / "No")
A4	list of strings	Cultural conflict descriptions (text explanations of conflicts in the image)

Refer to the actual data files in each task directory for exact schemas and usage.

---

Obtaining Manga109

Part of C³B is built on Manga109. For how to obtain the dataset (academic Manga109 or commercial subset Manga109-s), see the official site: Manga109 download and Manga109-s. The dataset is also available on Hugging Face: hal-utokyo/Manga109, hal-utokyo/Manga109-s.

---

Citation

If you use C³B, please cite:

@inproceedings{song2026c3b,
  title={Culture In a Frame: C³B as a Comic-Based Benchmark for Multimodal Cultural Awareness},
  author={Song, Yuchen and Chen, Andong and Zhu, Wenxin and Chen, Kehai and Bai, Xuefeng and Yang, Muyun and Zhao, Tiejun},
  booktitle={International Conference on Learning Representations (ICLR)},
  year={2026},
  url={https://openreview.net/forum?id=jvPdTOSTVl}
}

When using Manga109 (or Manga109-s) data, please also cite the recommended papers:

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

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

---

License

C³B

C³B (this benchmark, including annotations and task data in this repository) is released under CC BY 4.0 (Creative Commons Attribution 4.0 International). You may share and adapt the material with appropriate credit and under the same license terms.

Manga109 / Manga109-s

• Manga109: Permitted only for academic use by non-profit organizations. Redistribution to third parties is not permitted. Use is subject to the terms on the official Manga109 site; the University of Tokyo and the manga authors are not liable for use of the dataset.
• Manga109-s: A subset of 87 titles with permission for use by commercial organizations under the conditions on the Manga109-s page (e.g., limits on publishing whole pages, no redistribution, proper attribution). All other terms and disclaimers apply as stated there.

Use of Manga109 or Manga109-s is governed solely by the University of Tokyo laboratory and the official Manga109 website; users must comply with those terms and assume responsibility accordingly.