vis4d.eval.nuscenes

NuScenes evaluator.

class NuScenesDet3DEvaluator(data_root, version, split, save_only=False, class_map=None, metadata=('use_camera',), use_default_attr=False, velocity_thres=1.0)[source]

NuScenes 3D detection evaluation class.

__init__(data_root, version, split, save_only=False, class_map=None, metadata=('use_camera',), use_default_attr=False, velocity_thres=1.0)[source]: Initialize NuScenes evaluator.

__repr__()[source]

Concise representation of the dataset evaluator.

Return type:: str

property metrics: list[str]: Supported metrics.

gather(gather_func)[source]

Gather variables in case of distributed setting (if needed).

Parameters:: gather_func (Callable[[Any], Any]) – Gather function.
Return type:: None

reset()[source]

Reset evaluator.

Return type:: None

get_attributes(name, velocity)[source]

Get nuScenes attributes.

Return type:: str

process_batch(tokens, boxes_3d, velocities, class_ids, scores_3d, attributes=None)[source]

Process the results.

Return type:: None

evaluate(metric)[source]

Evaluate the results.

Return type:: tuple[Dict[str, Union[float, int, Tensor]], str]

save(metric, output_dir)[source]

Save the results to json files.

Return type:: None

class NuScenesTrack3DEvaluator(metadata=('use_camera',))[source]

NuScenes 3D tracking evaluation class.

__init__(metadata=('use_camera',))[source]: Initialize NuScenes evaluator.