VideoBadminton
收藏资源简介:
VideoBadminton是由奥本大学与国立中央大学联合创建的一个专注于羽毛球运动的高质量视频数据集。该数据集包含来自国立中央大学校队的19名男女运动员的羽毛球视频数据,涵盖了18种羽毛球动作,共7822个视频片段,总时长为145分钟。其构建严格遵循羽毛球世界联合会(BWF)的规范,确保了动作类别的全面性与专业性。此外,数据集还对球拍轨迹和球员位置进行了详尽标注。该数据集提供了丰富的动作类别和更细致的动作划分,旨在推动动作识别算法的发展,尤其是在理解和区分细微动作差异方面。
VideoBadminton is a high-quality video dataset focused on badminton, jointly developed by Auburn University and National Central University. This dataset includes badminton video data from 19 male and female athletes of the National Central University varsity team, covering 18 badminton actions, with a total of 7822 video clips and an overall duration of 145 minutes. Its construction strictly adheres to the regulations of the Badminton World Federation (BWF), ensuring the comprehensiveness and professionalism of the action categories. Furthermore, the dataset provides detailed annotations for racket trajectories and player positions. Boasting abundant action categories and finer-grained action partitioning, this dataset aims to advance the development of action recognition algorithms, particularly in understanding and differentiating subtle differences between actions.
数据集概述
- 数据集名称: VideoBadminton
- 数据集链接: VideoBadminton
模型性能
| 方法 | 骨干网络 | Top1准确率 | Top5准确率 | 平均分类准确率 | 模型链接 |
|---|---|---|---|---|---|
| R(2+1)D | ResNet2Plus1d | 79.53% | 96.11% | 66.97% | R(2+1)D |
| SlowFast | ResNet3dSlowFast | 82.80% | 97.54% | 73.80% | SlowFast |
| TimeSformer | TimeSformer | 73.18% | 94.78% | 57.70% | TimeSformer |
| Swim | SwinTransformer3D | 81.99% | 96.52% | 69.93% | Swim |
| MViT-V2 | MViT | 14.23% | 62.23% | 10.76% | MViT-V2 |
| ST-GCN | RecognizerGCN | 74.41% | 93.76% | 61.44% | ST-GCN |
| PoseC3D | Recognizer3D | 80.76% | 96.01% | 67.18% | PoseC3D |
数据集分割




