Manga109-v2026: Revisiting Manga109 Annotations for Modern Manga Understanding

The article announces the release of Manga109-v2026, an updated version of the Manga109 dataset, which addresses inaccuracies in dialogue text annotations identified in the original dataset. Key improvements include the revision of approximately 29,000 dialogue annotations through a combination of OCR-based issue detection and manual revision, correcting issues such as inaccurate transcriptions and overlapping dialogue. This update is significant for AI practitioners as it enhances the dataset's alignment with contemporary OCR and multimodal understanding tasks in manga, facilitating more effective training and evaluation of AI models in this domain.