JackAILab/OpenUni

OpenUni是一个大规模多模态视频数据集，旨在支持世界感知的视频生成和理解任务。该数据集包含四个主要子集：Koala36M（约500K视频）、HumanSingle（约400K视频）、HumanMulti（约100K视频）和OpenS2V（约300K视频），总计约3.9TB。数据集支持多种模态，包括RGB视频、深度图、光流、骨架检测、分割和密集姿态等。OpenUni是UnityVideo框架的官方训练数据集，适用于多种视频生成和估计任务。数据集还提供了灵活的加载选项，支持按需下载特定模态，以节省存储空间和时间。

OpenUni is a large-scale multi-modal video dataset designed for world-aware video generation and understanding. It consists of four major subsets: Koala36M (~500K videos), HumanSingle (~400K videos), HumanMulti (~100K videos), and OpenS2V (~300K videos), totaling approximately 3.9TB. The dataset supports various modalities including RGB videos, depth maps, optical flow, skeleton detection, segmentation, and dense pose. OpenUni serves as the official training dataset for the UnityVideo framework, which handles multiple video generation and estimation tasks. The dataset offers flexible loading options, allowing selective downloads of specific modalities to save storage space and time.