Motivation

Working with our own data has brought up the need to interact with the annotations programmatically. The annotation data is stored in .json files in the ASAM OpenLABEL annotation format, an emerging industry standard targeted towards the automotive sector. But as a standard it is designed very inclusively, which makes it overloaded for our limited use cases. We therefore decided to create a submodel called “RailLabel” with a corresponding devkit for easier interaction with the data. The example below shows a comparison between a purely JSON based approach compared to the devkit.

With JSON only:

import json

with open('path/to/file.json', 'r') as data_file:
    scene = json.load(data_file)

scene['openlabel']['frames']['0']['frame_properties']['streams']['lidar']['stream_properties']['stream_sync']['timestamp'] += 37

with open('path/to/other_file.json', 'w') as data_file:
    json.dump(scene, data_file)

With RailLabel:

import raillabel

scene = raillabel.load('path/to/file.json')
scene.frames['0'].streams['lidar'].timestamp += 37
raillabel.save(scene, 'path/to/other_file.json')

The devkit also includes other functionality like validating data against a schema and filtering the annotations.