brl-xfact/blv_eye4b
收藏Hugging Face2025-03-29 更新2025-11-01 收录
下载链接:
https://hf-mirror.com/datasets/brl-xfact/blv_eye4b
下载链接
链接失效反馈官方服务:
资源简介:
---
dataset_info:
features:
- name: image
dtype: image
- name: request
dtype: string
- name: response
dtype: string
- name: overall_score
dtype: float64
- name: avg_score
dtype: float64
splits:
- name: test
num_bytes: 141712566.0
num_examples: 98
download_size: 22285968
dataset_size: 141712566.0
configs:
- config_name: default
data_files:
- split: test
path: data/test-*
---
## Eye4B Dataset
📃 Can LVLMs and Automatic Metrics Capture Underlying Preferences of Blind and Low-Vision Individuals for Navigational Aid? [](https://arxiv.org/abs/2502.14883)
😀 We are excited to release the Eye4B dataset, a human-validated image-request-LVLM response data annotated with BLV preference scores!
This evaluation dataset can be used to assess (1) LVLM responses and (2) automatic metrics from perspectives of BLV preference alignment.
## Bibtex
```bash
@misc{an2025lvlmsautomaticmetricscapture,
title={Can LVLMs and Automatic Metrics Capture Underlying Preferences of Blind and Low-Vision Individuals for Navigational Aid?},
author={Na Min An and Eunki Kim and Wan Ju Kang and Sangryul Kim and Hyunjung Shim and James Thorne},
year={2025},
eprint={2502.14883},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2502.14883},
}
```
数据集信息:
特征:
- 名称:图像(image),数据类型:图像(image)
- 名称:请求文本(request),数据类型:字符串(string)
- 名称:回复文本(response),数据类型:字符串(string)
- 名称:整体评分(overall_score),数据类型:双精度浮点数(float64)
- 名称:平均评分(avg_score),数据类型:双精度浮点数(float64)
数据集划分:
- 名称:测试集(test),占用字节数:141712566.0,样本数量:98
下载大小:22285968 字节
数据集总大小:141712566.0 字节
配置项:
- 配置名称:默认(default),数据文件:
- 划分:测试集(test),路径:data/test-*
# Eye4B 数据集
📃 《LVLMs与自动指标能否捕捉视障与低视力人群对导航辅助工具的潜在偏好?》[]("https://arxiv.org/abs/2502.14883")
😀 我们很高兴发布Eye4B数据集——这是一项经过人工验证的图像-请求-大视觉语言模型(Large Vision Language Model,LVLM)回复数据集,标注了视障与低视力(Blind and Low-Vision,BLV)偏好评分!
本评估数据集可用于从两大维度开展评估:(1) 大视觉语言模型回复质量,(2) 自动评估指标与视障与低视力人群偏好的对齐程度。
## Bibtex
bash
@misc{an2025lvlmsautomaticmetricscapture,
title={Can LVLMs and Automatic Metrics Capture Underlying Preferences of Blind and Low-Vision Individuals for Navigational Aid?},
author={Na Min An and Eunki Kim and Wan Ju Kang and Sangryul Kim and Hyunjung Shim and James Thorne},
year={2025},
eprint={2502.14883},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2502.14883},
}
提供机构:
brl-xfact



