Argoverse2¶

Argoverse2 dataset management module aiming to access sensor data through log ids.

city:

"ATX": "Austin, Texas"
"DTW": "Detroit, Michigan"
"MIA": "Miami, Florida"
"PAO": "Palo Alto, California"
"PIT": "Pittsburgh, PA"
"WDC": "Washington, DC"

city origin:

"ATX": (30.27464237939507, -97.7404457407424)
"DTW": (42.29993066912924, -83.17555750783717)
"MIA": (25.77452579915163, -80.19656914449405)
"PAO": (37.416065,         -122.13571963362166)
"PIT": (40.44177902989321, -80.01294377242584)
"WDC": (38.889377,         -77.0355047439081)

Functions¶

Module Level¶

scene_point_etk.argoverse2.set_sensor_root(sensor_root=None, overwrite=True)¶

Module-level function to initialize and setup the Argoverse2 sensor dataset root directory and log id mapping.

Directory structure example:

<sensor_root>
    ├── 00a6ffc1-6ce9-3bc3-a060-6006e9893a1a  <- log_id
    │   ├── annotations.feather
    │   ├── calibration
    │   ├── city_SE3_egovehicle.feather
    │   ├── map
    │   └── ... (1 more files)
    ├── 01bb304d-7bd8-35f8-bbef-7086b688e35e
    ├── 022af476-9937-3e70-be52-f65420d52703
    └── ...

Parameters:

sensor_root (str | None, optional):
- Root directory of the Argoverse2 dataset. If None, the value from scene_point_etk/config.json is used.
overwrite (bool | True, optional):
- If True, update scene_point_etk/config.json with the provided sensor_root.

Log IDs¶

scene_point_etk.argoverse2.check_log_id(log_id)¶

Check if the log id can be located by this module.

For Example:

>> import scene_point_etk.argoverse2 as av2

# the log id without '-'
>> log_id = "00a6ffc16ce93bc3a0606006e9893a1a"

>> av2.set_sensor_root(sensor_root="a correct directory")
>> av2.check_log_id(log_id) # no error

>> av2.set_sensor_root(sensor_root="an incorrect directory")
>> av2.check_log_id(log_id)
"ValueError: Log id 00a6ffc16ce93bc3a0606006e9893a1a not found"

>> log_id = "non_existing_log_id"
>> av2.set_sensor_root(sensor_root="a correct directory")
>> av2.check_log_id(log_id)
"ValueError: Log id non_existing_log_id not found"

scene_point_etk.argoverse2.list_log_ids()¶: list over all available log ids

scene_point_etk.argoverse2.list_log_ids_by_mode(mode='test')¶: list over all available log_ids for a category: ‘train’, ‘val’, ‘test’

scene_point_etk.argoverse2.get_city_from_log_id(log_id)¶

Cities¶

scene_point_etk.argoverse2.check_city(city)¶

scene_point_etk.argoverse2.list_cities()¶

scene_point_etk.argoverse2.get_log_ids_from_city(city)¶

Images and Cameras¶

scene_point_etk.argoverse2.list_cameras_by_log_id(log_id)¶

Lidar Sweeps¶

scene_point_etk.argoverse2.list_sweep_files_by_log_id(log_id)¶

Classes¶

Images and Cameras¶

class scene_point_etk.argoverse2.ImageSequence(log_id, camera='ring_front_center')¶

get_an_image(index)¶: get an image as a HxWx3 numpy array by an integer index.

property figsize¶: Return the figure size as (height, width).

property intrinsic¶: Return the intrinsic matrix as a 3x3 numpy array.

property extrinsic¶: Return the extrinsic matrices as a Nx4x4 numpy array.

resize(H, W)¶: Return a resized copy of the image sequence with new height H and width W.

get_a_depth_map(index, points, invalid_value=-1, min_distance=0.0, max_distance=inf, return_details=False)¶

Get a depth map as a HxW numpy array by an integer index and 3D points.

Args:

index (int):
- index of the image in the sequence.
points (np.ndarray):
- Nx3 array of 3D points in world coordinates.
invalid_value (float | nan):
- value to assign to pixels where no points project.
min_distance (float):
- minimum distance from the camera to consider a point valid.
max_distance (float):
- maximum distance from the camera to consider a point valid.
return_details (bool):
- whether to return additional details as a dictionary.

Returns:

depth_map (np.ndarray):
- HxW array representing the depth map.
details (dict, optional):
- additional details if return_details is True.

align_timestamps(timestamps)¶: Return a copy of the Timestamps with rows aligned to the given timestamps.

class scene_point_etk.argoverse2.CameraSequence(log_id, cameras=['ring_front_center', 'ring_front_left', 'ring_front_right', 'ring_rear_left', 'ring_rear_right', 'ring_side_left', 'ring_side_right', 'stereo_front_left', 'stereo_front_right'])¶

align_timestamps(timestamps)¶: Return a copy of the CameraSequence with cameras aligned to the given timestamps.

Lidar Sweeps¶

class scene_point_etk.argoverse2.Sweep(path, coordinate='map')¶

class scene_point_etk.argoverse2.SweepSequence(log_id, coordinate='map')¶

3D Annotations¶

class scene_point_etk.argoverse2.Annotations(log_id, coordinate='map')¶