360度全景数据集

1. 简介360+x 数据集是面向场景理解的多模态数据集，以独特全景视角区别于传统数据集，通过多视角、多模态数据捕捉多样化场景；包含 2152 个多模态视频（共 8579k 帧），由 360° 摄像机（464 个视频）和 Spectacles 眼镜摄像机（1688 个视频）拍摄，覆盖 17 个城市的 28 个场景类别（15 个室内、13 个室外，含艺术空间、自然景观等）；视频被分段为 1380 个约 10 秒的短片，总时长约 67.78 小时，提供高分辨率（全景 / 双筒视频 5760x2880、第三人称视频 1920x1080）和低分辨率（全景视频 2432x1216、其余同高分辨率）两个版本，配套 38 个动作实例标签及预提取特征，满足不同计算资源需求The 360+x Dataset is a multimodal dataset oriented towards scene understanding. Distinguished from traditional datasets by its unique panoramic perspective, it captures diverse scenes through multi-view and multimodal data. It contains 2152 multimodal videos (totaling 8,579,000 frames), captured by 360° cameras (464 videos) and Spectacles eyewear cameras (1688 videos). The dataset covers 28 scene categories across 17 cities, including 15 indoor and 13 outdoor categories such as art spaces and natural landscapes. The videos are segmented into 1380 short clips of approximately 10 seconds each, with a total duration of about 67.78 hours. Two resolution versions are provided: the high-resolution version has 5760x2880 for panoramic/binocular videos and 1920x1080 for third-person videos, while the low-resolution version features 2432x1216 for panoramic videos and the same specifications as the high-resolution variant for other video types. Furthermore, it is equipped with 38 action instance labels and pre-extracted features to accommodate different computing resource requirements.

搜集汇总

背景与挑战

背景概述

360度全景数据集（360+x数据集）是一个面向场景理解的多模态数据集，以全景视角为特色，包含2152个多模态视频（共8579k帧），覆盖17个城市的28个室内外场景类别，提供高低分辨率版本及动作标签、预提取特征。该数据集适用于动作识别、视角转换建模等多模态学习任务，支持VR、AR和自动驾驶等领域的算法开发。