dataeval.types.Track¶

class dataeval.types.Track¶

All observations of a single object track within one video sequence.

track_id¶

The unique target ID assigned by the dataset.

Type:¶

int

boxes¶

Shape (T, 4) float32 array of bounding boxes in [x1, y1, x2, y2] (xyxy) format, one row per observation.

Type:¶

NDArray[Any]

frame_indices¶

Shape (T,) int64 array of 0-based frame indices corresponding to each row in boxes.

Type:¶

NDArray[Any]

scores¶

Shape (T,) float32 confidence scores (1.0 for ground-truth tracks).

Type:¶

NDArray[Any]

labels¶

Shape (T,) int64 0-based class indices. In practice a track should be single-class, but the raw per-frame labels are preserved so any label inconsistencies in the data remain visible.

Type:¶

NDArray[Any]