DAQUAR
收藏arXiv2025-09-30 收录
下载链接：
https://www.mpi-inf.mpg.de/departments/computer-vision-and-multimodal-computing/research/vision-and-language/visual-turing-challenge/
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个挑战性很大的问答任务数据集，它基于现实世界中的图像，展现了真实的室内场景。这些问题以不受限制的自然语言句子形式呈现。此外，DAQUAR数据集包含了与自然语言理解、常识知识以及由于细粒度类别而产生的歧义相关的多种挑战。在规模上，该数据集中问题部分包含1088个不同的名词，答案部分包含803个，总共涉及1586个不同的名词。其任务类型为问答任务。
This is a highly challenging question answering (QA) task dataset built on real-world images that showcase authentic indoor scenes. All questions are formulated as unconstrained natural language sentences. Additionally, the DAQUAR dataset features diverse challenges spanning natural language understanding, commonsense knowledge, and ambiguities stemming from fine-grained categorization. Regarding its scale, the question subset contains 1,088 distinct nouns, the answer subset contains 803, with a total of 1,586 distinct nouns across both subsets. The core task of this dataset is question answering.
