3x3 basketball pose dataset
收藏资源简介:
3x3篮球姿态数据集是由名古屋大学信息学研究生院构建的,专门用于多目标跟踪研究。该数据集包含3817帧连续视频帧和22902个关键点,显著超越了DeepSportRadar篮球瞬间数据集,并包含许多严重的遮挡场景。数据集的创建过程结合了主动学习和手动标注,旨在提高模型在生成虚拟标记时的性能。该数据集主要应用于团队运动中的多目标跟踪,旨在解决球员遮挡和视觉相似性带来的挑战,提升跟踪算法的准确性和效率。
The 3x3 Basketball Pose Dataset was constructed by the Graduate School of Informatics, Nagoya University, exclusively for multi-object tracking research. This dataset contains 3,817 consecutive video frames and 22,902 keypoints, significantly surpassing the DeepSportRadar Basketball Instant Dataset, and includes numerous scenarios with severe occlusions. The dataset was developed through a combination of active learning and manual annotation, with the goal of improving model performance when generating virtual annotations. Primarily applied to multi-object tracking in team sports, this dataset aims to address the challenges posed by player occlusions and visual similarity, thereby enhancing the accuracy and efficiency of tracking algorithms.

- 1Enhanced Multi-Object Tracking Using Pose-based Virtual Markers in 3x3 Basketball名古屋大学信息学研究生院 · 2024年



