SLU-2K
收藏资源简介:
SLU-2K是由摩德纳和雷焦艾米利亚大学等研究机构构建的语义评估基准数据集,旨在通过问答形式系统评估手语翻译系统的语义理解能力。该数据集包含2,350个封闭式视频问答对,基于广泛使用的PHOENIX-2014T和CSL-Daily数据集构建,涵盖动作、位置、数字、对象、人物、时间和天气条件等七类语义类别,通过自动化流水线生成并经过多阶段过滤以确保质量。数据集主要应用于手语理解领域,旨在解决传统翻译评估中语义保真度不足的问题,为开发更精准的辅助技术提供语义层面的评估工具。
SLU-2K is a semantic evaluation benchmark dataset constructed by the University of Modena and Reggio Emilia and other research institutions. It aims to systematically evaluate the semantic understanding capability of sign language translation systems through question-answering formats. This dataset contains 2,350 closed-ended video question-answering pairs, and is constructed based on two widely used datasets, PHOENIX-2014T and CSL-Daily. It covers seven semantic categories including actions, locations, numbers, objects, persons, time and weather conditions. Generated via an automated pipeline, it has undergone multi-stage filtering to ensure data quality. Primarily applied in the field of sign language understanding, this dataset is designed to solve the problem of insufficient semantic fidelity in traditional translation evaluation, providing a semantic-level evaluation tool for the development of more precise assistive technologies.
SLU-2K 数据集概述
基本信息
- 数据集名称:SLU-2K
- 数据集类型:基于问题的语义评估基准
- 应用领域:手语翻译(Sign Language Translation, SLT)的语义评估
主要功能
- 用于手语翻译模型的语义理解能力测试
- 通过问答形式评估翻译结果的语义准确性
包含内容
- PHOENIX-2014T 基准测试:针对德语手语数据集的语义评估
- CSL-Daily 基准测试:针对中文手语数据集的语义评估
- 自动生成管道:可用于为其他手语翻译数据集生成语义问答基准测试




