Nuscenes data set summary

One . brief introduction

nuScenes Data sets are created by Motional（ Formerly known as nuTonomy） Self driving public large-scale data set developed by the team . Collected in Boston and Singapore 1000 A driving scene .

nuScenes Data sets are inspired by groundbreaking KITTI Data sets .nuScenes It is the first to provide a complete sensor kit from autonomous vehicle （6 A camera 、1 A lidar 、5 A radar 、GPS、IMU） A large data set of data . And KITTI comparison ,nuScenes contain 7 Object annotation more than times .

Two . Data collection

1. Scenario Planning

About... Were collected in Boston and Singapore 15 Hourly driving data . For complete nuScenes Data sets , Released from Boston harbor and Singapore One North、 Data of Queenstown and Dutch village .

2. Car settings

Use two Renault vehicles with the same sensor layout Zoe Cars drive in Boston and Singapore . Data is collected from research platforms .

The camera （CAM） There are six , Separately distributed in the front （Front）、 Right front （Front Right）、 Left front （Front Left）、 The rear （Back）、 Right rear （Back Right）、 Rear left （Back Left）; Laser radar （LIDAR） Yes 1 individual , Place on the roof （TOP）; Millimeter wave radar has five , Place them separately in front （Front）、 Right front （Front Right）、 Left front （Front Left）、 Right rear （Back Right）、 Rear left （Back Left）.

3、 ... and . Data set download

1. To Nuscenes Download the dataset on the official website , Download address Poke it here

2. If you don't have an account, you can register , Because downloading data sets requires an account , Then choose this mini Version is OK

Four . Use of data sets

1. Import nuscenes-devkit library

pip install nuscenes-devkit

2. Load dataset information

from nuscenes.nuscenes import NuScenes
nusc = NuScenes(version='v1.0-mini', dataroot=' The specific path of the dataset ', verbose=True)

3. scene scene

Use the following code , View all scenarios in the current dataset

nusc.list_scenes()

Use the following code to view the information of a specific scenario

my_scene = nusc.scene[0]
print(my_scene)

4. sample sample

Every scene It lasts about 20s, that sample It's every 0.5 One sample per second . Or another way to think about it sample and scene,sence amount to 20s In the video ,sample It's every 0.5s Take a frame of image .

Use the following code , Get a specific one of the specific scenes sample Of token value

first_sample_token = my_scene['first_sample_token']  # Get the first one sample Of token value 
print(first_sample_token)

Use the following code , adopt sample Of token Values obtained sample Specific information

my_sample = nusc.get('sample', first_sample_token)
print(my_sample)

5. Sample data sample_data

Visualize the millimeter wave radar sensor in front

Through the following code , obtain sample Of data data

print(my_sample['data'])

(1) Get specific sample Information of specific sensors in

sensor_radar = 'RADAR_FRONT'  # The sensor selected here is the front millimeter wave radar sensor 
radar_front_data = nusc.get('sample_data',my_sample['data'][sensor_radar])
print(radar_front_data)

(2) Through sensor information token Value for visualization

nusc.render_sample_data(radar_front_data['token'])

Visualize the camera in front

#  Visualize the camera in front 
sensor_CAM_FRONT = 'CAM_FRONT'  # The sensor selected here is the front millimeter wave radar sensor 
CAM_FRONT_data = nusc.get('sample_data', my_sample['data'][sensor_CAM_FRONT])
print(CAM_FRONT_data)
nusc.render_sample_data(CAM_FRONT_data['token'])

Visual top lidar

#  Visual top lidar 
sensor_LIDAR_TOP = 'LIDAR_TOP'  # The sensor selected here is the front millimeter wave radar sensor 
LIDAR_TOP_data = nusc.get('sample_data', my_sample['data'][sensor_LIDAR_TOP])
print(LIDAR_TOP_data)
nusc.render_sample_data(LIDAR_TOP_data['token'])

Visualize other sensors in the same way ！！！

6. Sample mark sample_annotation

stay sample_data The information collected by the sensor has been shown in , This section will show the information marked on the sample , The method is similar to that before .

(1) obtain sample The label data of , And then output relevant information

my_annotation_token = my_sample['anns'][18]
my_annotation_metadata = nusc.get('sample_annotation',my_annotation_token)
my_annotation_metadata

(2) adopt sample Labeling information token value , Visualizing

#  Get specific sample Label information for 
my_annotation_token = my_sample['anns'][18]
my_annotation_metadata = nusc.get('sample_annotation', my_annotation_token)
print(my_annotation_metadata)
#  visualization 
nusc.render_annotation(my_annotation_metadata['token'])
time.sleep(3)

7. example instance

adopt nusc.instance[0] Get specific instances ,instance Represents an instance object , For example, a car

#  Get an instance object , And output its information 
my_instance = nusc.instance[5]
print(my_instance)
#  By way of example token value , Visualizing 
instance_token = my_instance['token']
nusc.render_instance(instance_token)
time.sleep(3)

8. Category categories

Use the following code to show all kinds in the dataset

nusc.list_categories()

9. attribute attributes

Show the specific attributes of the data set through the following code

nusc.list_attributes()

10. visualization visibility

#  Select the current sample One of the annotation information token value 
anntoken = my_sample['anns'][9]
nusc.render_annotation(anntoken)
time.sleep(3)

11. sensor sensor

The following code shows the sensor information in the data set

print(nusc.sensor)

12. Calibrate the sensor calibrated_sensor

Obtain the calibration information of the sensor through the following code

#  Show the calibration information of the sensor 
sensor_token = nusc.calibrated_sensor[0]
print(sensor_token)

13. Vehicle attitude ego_pose

Get the vehicle attitude information through the following code

#  Vehicle attitude ego_pose Information 
print(nusc.ego_pose[0])

14. journal log

Get the log information through the following code

#  Log information 
print(nusc.log[0])

15. Map map

#  Map information 
print(nusc.map[0])

The conclusion is over , Scatter flowers …