tags: CoachAI TrackNetV2 Dataset

Shuttlecock Trajectory Dataset

• Contributor: 林育慶、孫念恩、Tsì-Uí İk

Resource

• Download the dataset including professioanl matches and amateur games: https://nycu1-my.sharepoint.com/:u:/g/personal/tik_m365_nycu_edu_tw/EWisYhAiai9Ju7L-tQp0ykEBZJd9VQkKqsFrjcqqYIDP-g?e=S0AB1Z
• Labeling Tool: https://hackmd.io/CQmL6OKKSGKY9xUvU8n0iQ
• TrackNet : https://nol.cs.nctu.edu.tw:234/open-source/TrackNet
• TrackNetV2 : https://nol.cs.nctu.edu.tw:234/open-source/TrackNetv2

Introduction

Our dataset is created for the model training and testing of TrackNet and TrackNetV2 for badminton applications. The dataset is composed of 26 broadcast videos. The resolution and frame rate of videos are 1280×720 and 30 fps, respectively. A rally is the record starting from serving to its score. There are 78200 frames in total. Below are snapshots from the 23 videos. The first 23 of video (68675 frames) is from professional game, and the last 3 of video (9525 frames) is playing for fun by ourselves.

• Training Data (Professional)
  

• Training Data (Amateur)
  

• Testing Data
  

Descriptions of each video (train professional matches)

Rally of each match (train professional matchs)

Descriptions of each video (test professional matches)

Rally of each match (test professional matchs)

Folder Structure

Professional Game (match1 ~ match23)

​​​​ profession_dataset
​​​​        ├── match1/
​​​​        │     ├── ball_trajectory/
​​​​        │     │     ├── 1_01_00_ball.csv
​​​​        │     │     ├── 1_02_00_ball.csv
​​​​        │     │     ├── …
​​​​        │     │     └── *_**_**_ball.csv
​​​​        │     ├── frame/
​​​​        │     │     ├── 1_01_00/
​​​​        │     │     │     ├── 0.png
​​​​        │     │     │     ├── 1.png
​​​​        │     │     │     ├── …
​​​​        │     │     │     └── *.png
​​​​        │     │     ├── 1_02_00/
​​​​        │     │     │     ├── 0.png
​​​​        │     │     │     ├── 1.png
​​​​        │     │     │     ├── …
​​​​        │     │     │     └── *.png
​​​​        │     │     ├── …
​​​​        │     │     └── *_**_**/
​​​​        │     │
​​​​        │     └── rally_video/
​​​​        │           ├── 1_01_00.mp4
​​​​        │           ├── 1_02_00.mp4
​​​​        │           ├── …
​​​​        │           └── *_**_**.mp4
​​​​        ├── match2/
​​​​        │ ⋮
​​​​        └── match23/

Amateur Game (match1 ~ match3)

​​​​   amateur_dataset
​​​​        ├── match1/
​​​​        │     ├── ball_trajectory/
​​​​        │     │     ├── 1_00_01_ball.csv
​​​​        │     │     ├── 1_01_01_ball.csv
​​​​        │     │     ├── …
​​​​        │     │     └── *_**_**_ball.csv
​​​​        │     ├── frame/
​​​​        │     │     ├── 1_00_01/
​​​​        │     │     │     ├── 0.png
​​​​        │     │     │     ├── 1.png
​​​​        │     │     │     ├── …
​​​​        │     │     │     └── *.png
​​​​        │     │     ├── 1_01_01/
​​​​        │     │     │     ├── 0.png
​​​​        │     │     │     ├── 1.png
​​​​        │     │     │     ├── …
​​​​        │     │     │     └── *.png
​​​​        │     │     ├── …
​​​​        │     │     └── *_**_**/
​​​​        │     │
​​​​        │     └── rally_video/
​​​​        │           ├── 1_00_01.mp4
​​​​        │           ├── 1_01_01.mp4
​​​​        │           ├── …
​​​​        │           └── *_**_**.mp4
​​​​        ├── match2/
​​​​        │ 
​​​​        └── match3/

indexing of Rallies

In the dataset, a rally of a match is indexed by a set of three numbers s_ww_ll in which

• "s" is the set number.
• "ww" is the score of the match winner right after the rally.
• "ll" is the score of the match loser right after the rally.

For example, 1_04_02 denotes the rally in Set 1 after the rally the score goes to 4:2 and 4 is the score of the match winner and 2 is score of the match loser.

Naming of Files and Directories

Files and directories includes:

• readme.docx: the readme file of the dataset.
• match1, match2, …, match23: the match data directories
  • ball_trajectory: the folder for the trajectory csv files of rallies.
  • frame: the directory of frame images grouping by rallies
    • s_ww_ll: the directory to store the frames of the rally indexed by s_ww_ll
      • 0.png, 1.png, …: frame image in png format
  • rally_video: the directory to store the rally clips
    • s_ww_ll.mp4: the video clip of the rally s_ww_ll

Data Files

• In ball_trajectory folder, the labeling csv file is the csv with visibility column:

  

  The format is followed from labeling tool output.

  - Frame represent the frame of video. 
  - Visibility means the shuttlecock is visible or not at this frame(0 : invisible, 1 : visible).
  -  X, Y is the coordinate of shuttlecock.If shuttlecock is invisible now, then X, Y is 0.

• frame folder contains rally folder of each match :

  

  and In rally folder, there is the frame of each rally video(.png file) :

  

• In rally_video folder contains rally videos of match(.mp4 file):