Method __init__

backend/models/player_tracker.py:76–95

Initialize player tracker with a YOLOv8-Pose model. Using a pose model allows extracting 17 body keypoints per player per frame in the same inference pass — no extra compute cost. Args: model_path: Path to the YOLOv8-Pose model (e.g. 'yolov8m-pose.pt').

(self, model_path='yolov8m-pose.pt', device='cuda', imgsz: int = 1280, conf: float = 0.05)

class PlayerTracker:
    def __init__(self, model_path='yolov8m-pose.pt', device='cuda', imgsz: int = 1280, conf: float = 0.05):
        """
        Initialize player tracker with a YOLOv8-Pose model.

        Using a pose model allows extracting 17 body keypoints per player
        per frame in the same inference pass — no extra compute cost.

        Args:
            model_path: Path to the YOLOv8-Pose model (e.g. 'yolov8m-pose.pt').
            device: 'cuda' or 'cpu'
            imgsz: YOLO inference resolution. Should match input video resolution
                   to avoid downscaling small far-player detections below threshold.
            conf: Detection confidence threshold. Lower values surface more
                  candidates (needed for small far-court players).
        """
        self.model = YOLO(model_path)
        self.imgsz = imgsz
        self.conf = conf
        if device == 'cuda':
            self.model.to(device)

        # Dedicated detector for near-player recovery (predict-only). Kept
        # separate from self.model so its stateless predict() calls never