Manga109-v2026: Revisiting Manga109 Annotations for Modern Manga Understanding

Researchers have released Manga109-v2026, a revised version of the foundational Manga109 dataset, correcting roughly 29,000 dialogue annotations to better support modern optical character recognition and multimodal manga understanding systems [1][2]. The original Manga109 dataset, a widely used resource for AI research on Japanese comics, contained inaccurate transcriptions and coarse annotations that hindered its utility for contemporary tasks [1][2]. In a paper posted to the arXiv preprint server, authors led by Jeonghun Baek detail a systematic revision effort that identified five categories of annotation issues [1][2]. These include inaccurate transcriptions, missing text regions, overlapping dialogue and onomatopoeia, and under-segmented speech balloons [2]. The team combined automated detection using optical character recognition with manual correction to produce the updated dataset [2]. The first version of the paper was submitted on May 20, 2026, with a file size of 1,849 KB; a revised version followed on June 12, 2026, at 1,854 KB [1]. Manga109 has served as a benchmark as AI systems increasingly target manga understanding, translation, and related multimodal tasks [2]. The authors state that the revisions preserve the expressive structures characteristic of manga while improving alignment with modern systems [2]. The work appears on arXiv, an open-access repository for electronic preprints that, as of late 2024, receives about 24,000 submissions per month and hosts over two million articles [6]. The platform also supports community-built tools through its arXivLabs framework, which enables third-party developers to create features such as citation explorers and code finders that appear on article pages [4][5].