GUIDEDOG
收藏资源简介:
GUIDEDOG是一个面向盲人和低视力人群的辅助技术的新型可访问性感知指导数据集,包含22,084个现实世界场景图像描述对,其中包括2,106个由人类验证的数据。该数据集通过YouTube视频采样获得,覆盖了各种真实世界设置,每个场景都根据GUIDEDOG标准进行注释。此外,还构建了一个评估子集GUIDEDOGQA,包含多项选择题问答对,用于评估视觉感知能力。
GUIDEDOG is a novel accessibility-aware guidance dataset for assistive technologies targeting blind and low-vision populations. It contains 22,084 real-world scene image-caption pairs, among which 2,106 samples have been verified by human annotators. This dataset is sampled from YouTube videos, covering diverse real-world scenarios, with each scene annotated in accordance with the GUIDEDOG standard. Additionally, an evaluation subset named GUIDEDOGQA has been constructed, which consists of multiple-choice question-answer pairs for assessing visual perception capabilities.

- 1GuideDog: A Real-World Egocentric Multimodal Dataset for Blind and Low-Vision Accessibility-Aware GuidanceYonsei University · 2025年



