CoMix
收藏资源简介:
CoMix数据集由计算机视觉中心和MICC联合创建,是一个综合性的漫画理解基准,包含3.8k图像,涵盖多种漫画风格。数据集内容丰富,包括130K对象注释,30k文本-字符链接,以及33k字符集群。创建过程中,数据集通过精心选择和注释,确保了多样性和高质量。CoMix主要应用于漫画分析领域,旨在评估和提升模型在多任务处理和多模态推理方面的能力。
The CoMix dataset, jointly created by the Computer Vision Center and MICC, is a comprehensive benchmark for comic understanding. It includes 3.8k images spanning various comic art styles, and features rich annotation contents: 130k object annotations, 30k text-character alignments, and 33k character clusters. During its development, rigorous selection and annotation processes were employed to ensure both the diversity and high quality of the dataset. Primarily applied in the field of comic analysis, CoMix aims to evaluate and enhance the multi-task processing and multimodal reasoning capabilities of models.




