EmbodiedCity/UrbanVideo-Bench
收藏Hugging Face2025-07-15 更新2025-04-26 收录
下载链接：
https://hf-mirror.com/datasets/EmbodiedCity/UrbanVideo-Bench
下载链接
链接失效反馈官方服务：
资源简介：
UrbanVideo-Bench是一个用于评估视频大型语言模型（Video-LLMs）在处理连续的第一人称视觉观察方面的能力的数据集，这包括记忆、感知、推理和导航。数据集包含两部分：超过5000个多项选择题问答（MCQ）数据和超过1000个视频剪辑。MCQ数据存储在`MCQ.parquet`文件中，包含问题ID、视频ID、问题类别、问题和答案字段。
UrbanVideo-Bench is a dataset designed to evaluate the ability of video-large language models (Video-LLMs) to naturally process continuous first-person visual observations, including memory, perception, reasoning, and navigation. The dataset consists of two parts: over 5,000 multiple-choice question-answering (MCQ) data and over 1,000 video clips. The MCQ data is stored in the `MCQ.parquet` file, which includes fields for question ID, video ID, question category, question, and answer.
提供机构：
EmbodiedCity
