arXiv:2605.21182v2 Announce Type: replace-cross 
Abstract: Manga is a culturally distinctive multimodal medium and one of the most influential forms of Japanese popular culture. As AI systems increasingly target manga understanding, OCR, and translation, Manga109 has become a foundational dataset for manga-related AI research. However, the current Manga109 dataset contains inaccurate transcriptions and coarse annotations, which do not align well with modern OCR and multimodal manga understanding tasks. In this work, we revisit the dialogue text annotations of Manga109 and identify five categories of annotation issues, including inaccurate transcriptions, missing text regions, overlapping dialogue and onomatopoeia, and under-segmented speech balloons. To address these issues, we combine OCR-based issue detection and manual revision to construct Manga109-v2026, revising approximately 29,000 dialogue annotations. Our revisions better align Manga109 with modern OCR and multimodal manga understanding systems while preserving expressive structures characteristic of manga.

تم مراجعة مجموعة بيانات Manga109، وهي مورد حيوي لأبحاث الذكاء الاصطناعي في فهم المانغا، لمعالجة الأخطاء في تعليقاتها. النسخة المحدثة، Manga109-v2026، تصحح حوالي 29,000 تعليق حواري، مما يحسن التوافق مع مهام التعرف الضوئي على الحروف الحديثة والفهم متعدد الوسائط.

El conjunto de datos Manga109, un recurso crucial para la investigación en IA sobre la comprensión del manga, ha sido revisado para abordar inexactitudes en sus anotaciones. La versión actualizada, Manga109-v2026, corrige aproximadamente 29,000 anotaciones de diálogo, mejorando la alineación con las tareas modernas de OCR y comprensión multimodal.

Le jeu de données Manga109, une ressource essentielle pour la recherche en IA sur la compréhension des mangas, a été révisé pour corriger les inexactitudes dans ses annotations. La version mise à jour, Manga109-v2026, corrige environ 29 000 annotations de dialogue, améliorant l'alignement avec les tâches modernes de reconnaissance optique de caractères (OCR) et de compréhension multimodale.

The Manga109 dataset, a crucial resource for AI research in manga understanding, has been revised to address inaccuracies in its annotations. The updated version, Manga109-v2026, corrects approximately 29,000 dialogue annotations, improving alignment with modern OCR and multimodal understanding tasks.