OpenTTGames

包含38,752训练样本、9,502验证样本和7328测试样本的数据集。数据集的注释包括乒乓球的位置坐标、分割模板和游戏的状态信息。
OpenTTGames可用于与乒乓球相关的计算机视觉分析: 球检测，语义分割 (人，表，记分牌以及球坐标和游戏事件)。
该数据集包括用工业相机以120 fps录制的带注释的乒乓球比赛全高清视频。手动添加了注释，包括特定帧的帧号和相应的目标; 同时，手动注释了游戏中的事件，例如球弹跳，网击等; 以及球坐标和分段遮罩。

This dataset comprises 38,752 training samples, 9,502 validation samples, and 7,328 test samples. Its annotations include the position coordinates of table tennis balls, segmentation masks, and game state information. OpenTTGames can be utilized for table tennis-related computer vision analyses, such as ball detection and semantic segmentation (for humans, table, scoreboard, ball coordinates, and game events). This dataset includes annotated full-high-definition videos of table tennis matches recorded by industrial cameras at 120 fps. Annotations were manually added, covering the frame numbers of specific frames and their corresponding targets; meanwhile, game events during the matches like ball bounces, net hits, and other scenarios, as well as ball coordinates and segmentation masks were manually annotated.

搜集汇总

背景与挑战

背景概述

OpenTTGames是一个专注于乒乓球比赛的计算机视觉数据集，包含超过55,000个样本，提供球位置坐标、分割模板和游戏状态注释，适用于球检测和语义分割等任务。数据集基于120fps全高清视频录制，手动标注了球坐标、事件（如弹跳和网击）以及人、桌、记分牌的分割遮罩。由OSAI于2020年发布，支持乒乓球相关视觉分析研究。