Skip to content

initialize.md

Latest commit

 

History

History
209 lines (155 loc) · 4.72 KB

initialize.md

File metadata and controls

209 lines (155 loc) · 4.72 KB

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:

<!-- markdownlint-disable MD046 -->
```yaml
- 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:

<!-- markdownlint-disable MD046 -->
```yaml
- 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