Multi-Modal Experience Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/Aman-4-Real/MMTG
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是为了评估基于序列多模态信息的AI创作能力而全新手动标注的，包含了输入的图像-文本对以及来自歌曲歌词的输出句子。具体来说，该数据集中所包含的输出句子是从一个中文歌曲网站抓取的，而输入的图像-文本对则是从GraphMovie数据集中收集而来。这一任务的目的是基于多模态输入进行文本生成。
This dataset is a newly manually annotated collection developed to evaluate the AI creative capabilities based on sequential multimodal information. It contains input image-text pairs and output sentences derived from song lyrics. Specifically, the output sentences in this dataset were crawled from a Chinese music lyric website, while the input image-text pairs were collected from the GraphMovie Dataset. The goal of this task is text generation based on multimodal inputs.
提供机构：
Authors of the paper
