dataeval.extractors.DetectionGeometryExtractor¶

class dataeval.extractors.DetectionGeometryExtractor(model, *, confidence=0.0)¶

Turn a detection Model’s normalized boxes into per-detection geometry rows.

Each kept detection becomes [center_x, center_y, width, height, area, aspect], derived from normalized (x0, y0, x1, y1) boxes (resolution-independent). Detections whose max class score is below confidence are dropped.