YOLO Basketball Fine-Tuned Model
Overview
This model is a fine-tuned variant of the YOLO architecture, optimized for the detection of basketball-related entities in images.
It has been trained on a custom dataset composed of manually annotated images captured during a university-level basketball game.
The model is capable of identifying key elements within the court environment, including basketballs, players (with associated jersey colors and numbers), and referees. It is suitable for a range of downstream applications such as automated game analysis, player tracking, 3D scene reconstruction, and performance evaluation for coaching purposes.
Classes
The model supports detection of the following categories:
- Basketball
- Players, including jersey color and number annotations (e.g.,
red_23, blue_11)
- Referees (e.g.,
referee_1, referee_2)
The complete list of class labels can be accessed via the model.names attribute.
Training Details
- Base Model: YOLO26m
- Dataset: Custom basketball dataset with hand-annotated images
- Number of Epochs: 300
- Input Image Resolution: 1280 × 1280
The training pipeline and implementation details are available at the following repository:
https://github.com/446f6e6e79/player-tracking-in-sports/blob/main/finetune.ipynbUsage
1from ultralytics import YOLO
2
3model = YOLO("best.pt")
4results = model("image.jpg")
5results.show()
Notes
This model was fine-tuned on data from a single, specific basketball game.
As a result, it may exhibit little to no generalization to different teams, jersey styles, camera perspectives, or lighting conditions.
It is primarily intended for research and development purposes, and performance should be validated before deployment in broader or real-world scenarios.
