Dota2-Vis
收藏资源简介:
Dota2-Vis是由巴西天主教大学·巴拉那分校研究团队构建的计算机视觉数据集,专注于多人在线战术竞技游戏《Dota 2》的可见性分析。该数据集包含两个核心组成部分:一是基于2025年国际邀请赛全部144场职业比赛的双视角高清视频资源,共计288段1920×1080分辨率录像;二是2477张从职业比赛中提取并精细标注的小地图图像,涵盖玩家图标、克隆单位等21个标注类别。数据采集过程通过专业录制工具获取原始视频流,并采用人工标注流程对小地图元素进行边界框标注,确保在视觉密集场景下的标注质量。该数据集主要应用于游戏分析领域,旨在通过计算机视觉技术解决传统结构化数据难以捕捉的动态可见性问题,为战术策略分析、对手侦察和赛后复盘提供数据支持。
Dota2-Vis is a computer vision dataset constructed by the research team from the Pontifical Catholic University of Paraná, focusing on visibility analysis for the multiplayer online tactical competitive game *Dota 2*. The dataset comprises two core components: firstly, high-definition dual-perspective video resources sourced from all 144 professional matches of The International 2025, totaling 288 video clips with a resolution of 1920×1080; secondly, 2,477 mini-map images extracted from professional matches and meticulously annotated, covering 21 annotation categories such as player icons and cloned units. During the data collection process, raw video streams are acquired using professional recording tools, and manual annotation workflows are employed to perform bounding box annotation on mini-map elements, ensuring annotation quality in visually dense scenarios. This dataset is primarily applied in the field of game analysis, aiming to address dynamic visibility problems that traditional structured data is unable to capture through computer vision technologies, providing data support for tactical strategy analysis, opponent reconnaissance, and post-match review.




