nipsedtrack2026/q2-cubemap-mcq
收藏资源简介:
该数据集名为Q2 — 视野外物体定位(立方体地图多选),是一个基于自我中心视觉和厨房环境空间推理的视觉问答数据集。数据集包含基于Epic-Kitchens和HD-Epic数据集的立方体地图投影帧的多选题。设计用于视觉语言模型(VLM)的基准测试,并包含两种干扰项策略用于问题生成。数据集采用CC-BY-NC 4.0许可发布,供学术使用,详细描述了数据集结构、文件模式、渲染过程及评估指标。
The dataset, named Q2 — Out-of-View Object Localization (Cubemap Multiple Choice), is a visual-question-answering dataset focusing on egocentric vision and spatial reasoning in kitchen environments. It consists of multiple-choice questions based on cubemap projections of frames from the Epic-Kitchens and HD-Epic datasets. Designed for benchmarking visual language models (VLMs), it includes two distractor strategies for question generation. Released under the CC-BY-NC 4.0 license for academic use, the dataset provides comprehensive details on its structure, file schema, rendering process, and evaluation metrics.




