GRAFFORD
收藏arXiv2024-02-20 更新2024-06-21 收录
下载链接：
https://github.com/sayantan11995/Affordance
下载链接
链接失效反馈官方服务：
资源简介：
GRAFFORD数据集是由印度理工学院卡拉格普尔分校的研究团队创建，旨在评估语言模型和视觉语言模型对物体功能的理解能力。该数据集包含2368个句子-对象对，每个对都标注了15种功能类别，这些类别是从视觉和语言领域收集的。数据集的创建过程涉及从大规模未标注文本中学习上下文表示，并通过众包平台toloka.ai训练标注者。GRAFFORD数据集的应用领域包括自然语言理解任务，特别是评估模型对不常见物体功能的推理能力。
The GRAFFORD dataset was created by a research team from the Indian Institute of Technology Kharagpur, aiming to evaluate the capability of language models and vision-language models in understanding object functions. This dataset contains 2,368 sentence-object pairs, each annotated with 15 functional categories collected from both visual and linguistic domains. The construction of the GRAFFORD dataset involved learning contextual representations from large-scale unannotated text, and training annotators via the crowdsourcing platform toloka.ai. Application scenarios of the GRAFFORD dataset include natural language understanding tasks, particularly for evaluating models' reasoning abilities regarding the functions of uncommon objects.
提供机构：
印度理工学院，卡拉格普尔分校
创建时间：
2024-02-20
