COO

COO数据集是由东京大学创建的，专注于日本漫画中的拟声词文本识别。该数据集包含大量的任意形状和位置的文本，如极度弯曲或部分缩小的文本，以及分离成多个部分的截断文本。数据集的创建旨在推动对不规则文本识别的研究，特别是预测截断文本之间的链接，以捕捉其意图表达的意义。COO数据集适用于文本检测、文本识别和链接预测任务，旨在解决复杂场景下的文本识别问题，特别是在漫画分析和翻译领域具有重要应用价值。

The COO dataset was developed by The University of Tokyo, focusing on onomatopoeic text recognition in Japanese manga. This dataset contains a large number of texts with arbitrary shapes and positions, such as extremely curved or partially scaled-down texts, as well as truncated texts split into multiple segments. The dataset is designed to promote research on irregular text recognition, particularly predicting the links between truncated texts to capture their intended meaning. The COO dataset is applicable to text detection, text recognition and link prediction tasks, aiming to address text recognition challenges in complex scenarios, and holds significant application value especially in the fields of manga analysis and translation.