critique-VQA
收藏资源简介:
该数据集是一个用于问答任务的数据集,包含大约29012个训练样本,数据量在10K到100K之间。数据集的特征包括对话、选定的回答、被拒绝的回答和图像。每个对话、选定的回答和被拒绝的回答都包含'from'和'text'两个字段,图像字段则直接存储图像数据。数据集的配置名为'default',数据文件路径为'data/train-*'。数据集的下载大小为3545659131字节,总大小为9094031507.592字节。
This dataset is designed for question answering tasks, containing approximately 29,012 training samples, with a data volume ranging from 10K to 100K. Its features include dialogues, selected responses, rejected responses, and images. Each dialogue, selected response, and rejected response contains two fields: 'from' and 'text', while the image field stores image data directly. The dataset configuration is named 'default', and the data file path is 'data/train-*'. The download size of the dataset is 3545659131 bytes, and the total size is 9094031507.592 bytes.
数据集概述
基本信息
- 许可证: MIT
- 任务类别: 问答
- 数据规模: 10K<n<100K
配置
- 配置名称: default
- 数据文件:
- 分割: train
- 路径: data/train-*
- 数据文件:
数据集信息
-
特征:
- conversation:
- from: string
- text: string
- chosen:
- from: string
- text: string
- rejected:
- from: string
- text: string
- image: image
- conversation:
-
分割:
- train:
- 字节数: 9094031507.592
- 样本数: 29012
- train:
-
下载大小: 3545659131
-
数据集大小: 9094031507.592
引用
bibtex @article{zhang2024critic, title={Critic-V: VLM Critics Help Catch VLM Errors in Multimodal Reasoning}, author={Zhang, Di and Lei, Jingdi and Li, Junxian and Wang, Xunzhi and Liu, Yujie and Yang, Zonglin and Li, Jiatong and Wang, Weida and Yang, Suorong and Wu, Jianbo and others}, journal={arXiv preprint arXiv:2411.18203}, year={2024} }




