VideoCon
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/Hritikbansal/videocon
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为VideoCon，旨在通过利用大型语言模型生成的对比性字幕来微调视频-文本对齐模型，以提高其鲁棒性。该数据集包含视频片段、原始字幕、对比性字幕以及解释，这些内容被用于训练和评估视频-语言蕴含模型。其所涉及的任务是视频-语言对齐。
The dataset is named VideoCon, which aims to fine-tune video-text alignment models by leveraging contrastive captions generated by large language models to enhance their robustness. This dataset includes video clips, original captions, contrastive captions and explanations, which are used for training and evaluating video-language entailment models. The task involved in this dataset is video-language alignment.
提供机构：
Authors of the paper
