CatBoost: unbiased boosting with categorical features
Paper • 1706.09516 • Published
CatBoost Tennis Bounce Detector
A CatBoost gradient boosting model trained to detect ball bounces from kinematic features derived from ball tracking data.
Input Features
The model takes kinematic features computed from ball trajectory data, including:
- Ball position (x, y) over a sliding window
- Velocity and acceleration components
- Vertical direction changes (V-shape patterns)
- Frame-to-frame displacement
Usage
from catboost import CatBoostClassifier
model = CatBoostClassifier()
model.load_model("bounce_model.cbm")
predictions = model.predict(features)
Post-processing
Predictions are validated with:
- V-shaped Y-reversal confirmation
- Court-view status filtering
- Minimum inter-bounce gap enforcement
Credits
- CatBoost by Yandex. Paper: CatBoost: unbiased boosting with categorical features (NeurIPS 2018). Developed by Anna Veronika Dorogush, Vasily Ershov, Andrey Gulin.
- Bounce detection approach inspired by yastrebksv/TennisProject. Training data derived from ball trajectories extracted via TrackNet from TenniSet broadcast videos.
Part of
This model is part of the Sports Commentator from Video project — an automated tennis commentary pipeline using TrackNet, CatBoost, VideoMAE, Qwen2.5, RF-DETR, and F5-TTS.