XSpaceCoderX/AD-Rallies
收藏资源简介:
AD-Trajectories数据集是为硕士学位论文《从广播到3D:一种用于网球轨迹和旋转估计的深度学习方法》创建的,由奥格斯堡大学机器学习与计算机视觉系Alexandra Göppert制作。该数据集是一个大规模合成数据集,使用MuJoCo物理引擎生成,旨在通过提供高度准确的物理模型(如马格努斯效应和复杂的球与场地交互)来缩小合成与真实之间的差距。数据集包含约320万次合成的网球对打,从发球开始,最多可包含4次基本击球(如底线击球、截击、高球、短球和扣杀)。所有物理运动学数据,包括3D位置、线速度和角速度(旋转),均以500帧每秒的高分辨率记录,对应时间步长为0.002秒。数据以.npz文件形式存储,每个文件包含七个.npy文件,分别存储空间、时间和相机数据。数据集的总大小为87GB,解压后需要约180GB的磁盘空间。
The AD-Trajectories dataset was created for the Masters thesis "From Broadcast to 3D: A Deep Learning Approach for Tennis Trajectory and Spin Estimation" by Alexandra Göppert at the University Augsburg, Chair of Machine Learning and Computer Vision. The dataset is a large-scale synthetic dataset generated using the MuJoCo physics engine, designed to bridge the synthetic-to-real gap by providing highly accurate physical models of aerodynamic forces, such as the Magnus effect, and complex ball-court interactions. The dataset comprises approximately 3.2 million synthetic tennis rallies, starting with a ball toss and a serve, followed by up to 4 further basic strokes (e.g., groundstroke, volley, lob, short, and smash). All physical kinematics, including 3D positions, linear velocities, and angular velocities (spin), are captured at a high resolution of 500 frames per second (fps), corresponding to a time step of 0.002 seconds. The dataset is stored as .npz files, each containing seven .npy files that store spatial, temporal, and camera data. The total size of the dataset is 87 GB, and extracting it requires approximately 180 GB of disk space.




