Skip to content

Dataset Schema

Directory Structure

<DATASET_ID>/
└── <DATASET_VERSION>/
    ├── annotation/    ...schema tables in JSON format
    ├── data/          ...sensor raw data
    ├── input_bag/     ...original ROS bag file
    ├── map/           ...map files
    ├── lidarseg/      ...[OPTIONAL] LiDAR segmentation annotation
    └── status.json    ...dataset status information

Schema Tables

Annotation information is stored in the annotation directory.

See the document of dataset schema for details.

Sensor Data

Raw sensor data is stored in the data directory.

See the document of sensor data for details.

Map Data

Map data is stored in the map directory.

It is structured as follows:

map/
├── lanelet2_map.osm    ...lanelet2 map file
└── pointcloud_map.pcd/ ...pointcloud map directory

LiDAR Segmentation Annotation

T4 dataset can include 3D LiDAR segmentation annotation optionally. The format is exactly the same as the nuScenes format with one additional lidarseg.json file.

Note that every <lidarseg_token>.bin consists of category labels for every LiDAR pointcloud at a keyframe.

lidarseg/
└── annotation/
    ├── <lidarseg_token>.bin    ...category labels for every LiDAR pointcloud at a keyframe
    ...

status.json

status.json contains the information about the configuration used to generate the dataset.