CoachAI Badminton Challenge 2023
In conjunction with IJCAI 2023 (Macao, S.A.R)
@ August 19th-25th, 2023
Track 2: Forecasting Future Turn-Based Strokes in Badminton Rallies
Introduction
The goal of this track is to forecast future strokes including shot types and locations given the past stroke sequences, namely stroke forecasting. For each singles rally, given the observed 4 strokes with type-area pairs and two players, the goal is to predict the future strokes including shot types and area coordinates for the next n steps. n is various based on the length of the rally.
Problem Definition
Evaluation Metrics
Data
Input: landing_x, landing_y, shot type and metadata of past 4 strokes
Output: landing_x, landing_y, shot type of future strokes
Column meaning
train.csv, val.csv, test.csv
rally: serial number of rallies in a match
ball_round: the order of the shot in a rally
time (hr:min:sec): the shot’s hitting time
frame_num: sec * fps = frame_num
roundscore_A: Player A’s current score in the set
roundscore_B: Player B’s current score in the set
player: the player who performed the shot
type: the type of shot, 10 categories, i.e., short service, long service, clear, push/rush, smash, defensive shot, drive, net shot, lob, drop. (the naming might be different for the two tracks)
aroundhead: hit the shuttle around the head or not
backhand: hit the shuttle with back hand or not
landing_height: if the shuttle destinations is hit above (1) or below (0) the net
landing_area: the grid of the shuttle destinations
landing_x, landing_y: the coordinates of the shuttle destinations
lose_reason: the reason why the rally ended. To be capable of different usages, we record the timest
getpoint_player: the player who won the rally
player_location_area: the location of the player who performed the shot
player_location_x, player_location_y
opponent_location_area: the location of the player who prepared to receive the shot
opponent_location_x, opponent_location_y: the coordinates of the opponent’s location when the player hits the shuttle
set: the current set in a match (best-of-3)
match_id: serial number of matches
rally_id: serial number of all rallies
rally_length: number of shots in a rally, rally_length-4 is the shots needed to predict
sample_id: the predictions are required to sample 6 times for each instance, i.e., 1~6
match_metadata.csv
match_id: serial number of matches
set: the number of played sets in a match (either 2 or 3)
duration: the duration of a match in minutes
winner: the winner player
loser: the lose player
homography_matrix: Help transform coordinates from the real-world system back to the camera system by p=H-1p’
For getting access to the data, please fill in the form.