NuScenesDataExtractor

Data extractor class for extracting data from nuScenes dataset

Prerequisites

Download and set up nuScenes.

Run example

Specify nuScenes python-sdk path in main.py and nusc_data_extractor.py, then run:

>>> python main.py

Commands

Control

Select the desired scene with

scene_name = 'scene-0001'
de_nusc.set_scene(scene_name)

For iterating through the time steps of the scene use

de_nusc.advance()

It is also possible to get a list with all the scenes using

scene_list = de_nusc.get_scene_list()

Data extraction

Extract data of the lidar sensor for current time step:

lidar_points = de_nusc.get_lidar_pointcloud()

Extract data of one radar sensor for current time step:

radar_channel = 'RADAR_FRONT'
radar_points, radar_velocities = de_nusc.get_radar_pointcloud(radar_channel)

Extract data of all radar sensors for current time step:

radar_points, radar_velocities = de_nusc.get_all_radar_pointclouds()

Extract raw image of one camera sensor for current time step:

camera_channel = 'CAM_FRONT'
camera_image = de_nusc.get_camera_image(camera_channel)

Get annotations from the current scene using

bboxes, class_ID, velocity=de_nusc.get_annotations(1) #1 if we want 2D bboxes or 2 if we want 3D bboxes

Get annotations for all possible annotations. If a sensor channel (e.g. 'RADAR_FRONT') is specified, just the annotations seen by the sensor are given.

de_nusc.get_annotations()
de_nusc.get_annotations('CAM_FRONT')
de_nusc.get_annotations('RADAR_FRONT')

Saving the data

It is possible to save the current LiDAR point cloud into an image of a fixed maximum axis dimmension as well as the labels of that image using:

#Save the image with the LiDAR points, the maximum axis value and the name used to save the file. The labels will also be saved
de_nusc.save_figure_and_labels("LIDAR_TOP", 40, "image0000")

Use the code "labelme2coco" from: https://github.com/Tony607/labelme2coco/blob/master/labelme2coco.py to get COCO structure for the labels.

gaborba1 / NuScenesDataExtractor