Skip to content

Initialization

Initialize Tier4 class

Tier4 class expects both following dataset directly structure with or without <VERSION> directory:

  • With <VERSION> directory:
data/tier4/
└── <VERSION> ...version number
    ├── annotation ...contains `*.json` files
    ├── data
    │   ├── CAM_BACK
    │   ├── CAM_BACK_LEFT
    │   ├── CAM_BACK_RIGHT
    │   ├── CAM_FRONT
    │   ├── CAM_FRONT_LEFT
    │   ├── CAM_FRONT_RIGHT
    │   ├── LIDAR_CONCAT
    │   └── ...Other sensor channels
    ...
  • Without <VERSION> directory:
data/tier4/
├── annotation ...contains `*.json` files
├── data
│   ├── CAM_BACK
│   ├── CAM_BACK_LEFT
│   ├── CAM_BACK_RIGHT
│   ├── CAM_FRONT
│   ├── CAM_FRONT_LEFT
│   ├── CAM_FRONT_RIGHT
│   ├── LIDAR_CONCAT
│   └── ...Other sensor channels
...

You can initialize a Tier4 instance as follows:

>>> from t4_devkit import Tier4

>>> t4 = Tier4("data/tier4/", verbose=True)
======
Loading T4 tables in `annotation`...
Reverse indexing...
Done reverse indexing in 0.007 seconds.
======
13 attribute
7 calibrated_sensor
8 category
2524 ego_pose
106 instance
1 log
1 map
72 sample
2390 sample_annotation
2524 sample_data
4 visibility
7 sensor
1 scene
1326 object_ann
0 surface_ann
0 keypoint
Done loading in 0.061 seconds.
======

Note that if you doesn't specify revision parameter in construction, it searches the latest version of the dataset. By specifying revision=<VERSION>, you can load the specific version of the dataset.

>>> t4 = Tier4("data/tier4/", revision="2", verbose=True)

Access to Schema Fields

scene

>>> my_scene = t4.scene[0]

sample

sample is an annotated keyframe of a scene at a given timestamp.

>>> first_sample_token = my_scene.first_sample_token
>>> my_sample = t4.get("sample", first_sample_token)

You can access the sample_data associated with this sample. sample.data returns a dict object consists of {str: <SAMPLE_DATA_TOKEN (str)>}.

>>> my_sample.data

sample_data

sample_data is references to a family of data that is collected from specific sensors.

>>> sensor = "CAM_FRONT"
>>> t4.get("sample_data", my_sample.data[sensor])

sample_annotation

sample_annotation refers to any 3D bounding box in a corresponding sample. All location data is given with respect to the global coordinate system. You can access the list of sample_annotation tokens with sample.ann_3ds: list[str].

>>> my_annotation_token = my_sample.ann_3ds[0]
>>> t4.get("sample_annotation", my_annotation_token)

instance

Each annotated object is instanced to be tracked.

>>> t4.instance

category

A category is the object assignment of an annotation.

>>> t4.category

attribute

An attribute is a property of an instance that may change throughout different parts of a scene while category remains the same.

>>> t4.attribute

visibility

visibility is defined as the fraction of pixels of a particular annotation that are visible over the 6 camera feeds.

>>> t4.visibility
Warning

Expected level values in visibility are as below:

- full    : No occlusion for the object.
- most    : Object is occluded, but by less than 50%.
- partial : Object is occluded, but by more than 50%.
- none    : Object is 90-100% occluded and no points/pixels are visible.

Following old formats are also supported but deprecated:

- v80-100 : full
- v60-80  : most
- v40-60  : partial
- v0-40   : none

If input level does not satisfy any above cases, VisibilityLevel.UNAVAILABLE will be assigned.

sensor

T4 dataset consists of several type of sensors. The supported sensor modalities and channels are defined in t4_devkit/schema/tables/sensor.py.

>>> t4.sensor

calibrated_sensor

calibrated_sensor consists of the definition of a calibration of a particular sensor based on a vehicle.

>>> t4.calibrated_sensor

Note that the translation and rotation parameters are given with respect to the ego vehicle body frame.

ego_pose

ego_pose contains information about the translation and rotation of the ego vehicle, with respect to the global coordinate system.

>>> t4.ego_pose