Matrix-Game-MC

该数据集是一个全面性的Minecraft游戏数据集，包含了超过2,700小时未经标注的游戏视频片段，以及超过1,000小时的高质量标注视频片段。这些标注细致到了键盘和鼠标操作的级别。数据集涵盖了多样的生物群落，并经过严格的筛选过程，确保了视频片段的高质量和信息丰富性，非常适合用于模型训练。规模上，该数据集拥有超过2,700小时的未标注视频和1,000小时的标注视频。其任务领域涉及互动式视频生成和世界建模。

This is a comprehensive Minecraft gameplay dataset, containing over 2,700 hours of unannotated gameplay video clips and over 1,000 hours of high-quality annotated video clips, with annotations detailed down to the granularity of keyboard and mouse inputs. Covering a diverse range of in-game biomes, the dataset has undergone a rigorous screening process to ensure the high quality and informational richness of the video clips, making it highly suitable for machine learning model training. In terms of scale, the dataset includes over 2,700 hours of unannotated videos and 1,000 hours of annotated videos. Its applicable task domains involve interactive video generation and world modeling.

搜集汇总

背景与挑战

背景概述

Matrix-Game-MC是Skywork AI发布的开源世界模型系列，包含三个版本，专注于实时交互式视频生成和长时记忆功能。该项目采用MIT许可证，具有较高的社区关注度（2.2k stars）。