CoachAI Badminton Challenge 2023
In conjunction with IJCAI 2023 (Macao, S.A.R)
@ August 19th-25th, 2023
Track 1: Automatic Annotation of Technical Data for Badminton Match Videos
Introduction
In this track, participants develop computer vision technology to automatically extract shot-by-shot technical data from the broadcast video of badminton matches. The shot-by-shot data include the temporal, spatial, posture, and skill information as the shuttlecock is hit during the match such as data at the hitting time, ball position, hitting player, swing posture, standing position of both sides, used ball type, and winner. Those are composed of the big data for technical and tactical analysis of badminton matches. The scoring is rally-based, where each rally is evaluated separately and can get up to 1 point in total according to the correctness of the number of shots in the rally and the accuracy of the predicted attributes. The final ranking is sorted from highest to lowest according to the total score, with the highest being the first and the lowest being the last.
Evaluation
Each single video clip is scored separately, with a total score of 1 point per video clip. First, the number of shots in the rally video is validated, and each row of the CSV file is considered as one shot.Â
If the predicted number of shots is different from the ground truth, the question will be scored as 0 points.
If the number of shots is the same, 0.1 points will be awarded, and proceed to the next stage to compare the content of the columns, with the highest score for column content being 0.9 points.
If the number of shots is correct, compare each shot in sequence, with each shot scored according to the total score of the columns, and the average of each shot's score as the score for content. First, the HitFrame column is validated. If the prediction error is greater than 2 frames compared to the true value, the shot will get 0 points in content and proceed to the next shot for scoring; if the prediction error is not greater than 2 frames, 0.1 points will be given, and proceed to compare other columns in the same shot.
Hitter: 0.1 points if correct, 0 points otherwise.
BallHeight: 0.1 points if correct, 0 points otherwise.
Landing: If the prediction error is not greater than 6 pixels in the Euclidean distance, it is considered correct and is awarded 0.1 points; otherwise, 0 points.
HitterLocation: If the prediction error is not greater than 10 pixels in the Euclidean distance, it is considered correct with 0.05 points, otherwise 0 points.
DefenderLocation: If the prediction error is not greater than 10 pixels in the Euclidean distance, it is considered correct with 0.05 points, otherwise 0 points.
Backhand: 0.05 points if correct, 0 points otherwise.
RoundHead: 0.05 points if correct, 0 points otherwise.
BallType: 0.2 points if correct, 0 points otherwise.
Winner: 0.1 points if correct, 0 points otherwise. (Note: If it is not the last shot but is filled in, it will be considered incorrect.)
Assuming that there are R rally videos in the data set, and the i-th video has Si shots, the scoring formula is given below:
ASSi (the Average Shot Score of the content of the i-th rally video) is given by
with SSj (the j-th Shot Score) given by:
Data
This dataset consists of international competition videos organized by the Badminton World Federation (BWF). The recorded matches are shot from the back of and above the players. Each video will record a rally from the serve to the dead ball stage and the marked file after the shot mark, with an mp4 encoding and a CSV format for the marked file, and a frame rate of 30 FPS and a resolution of 1280*720. The dataset includes 13 annotation fields. The following are the descriptions of each annotation field:
VideoName: The file name of each rally video
HitFrame: The moment of the shot in the rally, in frames
Hitter: The player hitting the ball, marked with A/B
A: Sited farther from the camera
B: Sited closer to the camera
BallHeight: The height of the hitting point, marked by 1/2
1: The hitting point is above the net
2: The hitting point is below the net
RoundHead: Whether the player has a round-head hitting action, marked by 1/2
1: Round-head hitting
2: No round-head hitting
Backhand: Whether the player's hitting action is backhand, marked by 1/2
1: Backhand hitting
2: Not backhand hitting
LandingX/LandingY: The projection on the court plane of the shuttlecock at the end of the trajectory either hit by the opponent or touch the ground, in pixels
HitterLocationX/HitterLocationY: The location of the striking player, taking the toe of the habit foot as the coordinates in the picture, in pixels. If there is a jump, the coordinates projected onto the court plane are the answer
DefenderLocationX/DefenderLocationY: The location of the defending player, similar to the location of the hitting player
BallType: Nine types of balls, marked by 1~9. 1~2 are the types of serves, used only for serving (or first shot); 3~9 are the types of returns, used for subsequent shots
1: Short serve
2: Long serve
3: Clear
4: Drive
5: Smash
6: Net shot
7: Drop shot
8: Lift
9: Push shot
Winner: Winning and losing points, divided into three categories, marked with A/B/X. Only the last shot of each rally video can be scored, and the rest are marked with X
A: Score by upper half-court player
B: Score by lower half-court player
X: No score
For getting access to the data, please fill in the form.
Example