cyber_record

cyber_record is a cyber record file offline parse tool. You can use cyber_record to read messages from record file, or write messages to the record file.

os	support	remark
ubuntu	✔️
mac	✔️
windows	✔️

Quick start

First install "cyber_record" by the following command.

pip3 install cyber_record
// or update version
pip3 install cyber_record -U

Since the high-level protobuf version needs to regenerate the proto_pb file, we limit protobuf version to 3.19.4

Demo record

You can download a apollo demo record from demo_sensor_data_for_vision

Command line mode

You can easily get the information in the record file by the following command.

Info

cyber_record info will output the statistics of the record file.

$ cyber_record info -f example.record.00000

record_file: example.record.00000
version:     1.0
begin_time:  2021-07-23 17:12:15.114944
end_time:    2021-07-23 17:12:15.253911
duration:    0.14 s
size:        477.55 KByte
message_number: 34
channel_number: 8

/apollo/planning                      , apollo.planning.ADCTrajectory         , 1
/apollo/routing_request               , apollo.routing.RoutingRequest         , 0
/apollo/monitor                       , apollo.common.monitor.MonitorMessage  , 0
/apollo/routing_response              , apollo.routing.RoutingResponse        , 0
/apollo/routing_response_history      , apollo.routing.RoutingResponse        , 1
/apollo/localization/pose             , apollo.localization.LocalizationEstimate, 15
/apollo/canbus/chassis                , apollo.canbus.Chassis                 , 15
/apollo/prediction                    , apollo.prediction.PredictionObstacles , 2

Echo

cyber_record echo will print the message of the specified topic to the terminal.

$ cyber_record echo -f example.record.00000 -t /apollo/canbus/chassis

engine_started: true
speed_mps: 0.0
throttle_percentage: 0.0
brake_percentage: 0.0
driving_mode: COMPLETE_AUTO_DRIVE
gear_location: GEAR_DRIVE
header {
  timestamp_sec: 1627031535.112813
  module_name: "SimControl"
  sequence_num: 76636
}

Or you can reference the cyber_record in the python file by

from cyber_record.record import Record

Recover

If you find record file is missing index, you can recover the file by cyber_record recover.

It is best to backup the file before recover!!!

1. Generate the file descriptor set. Must be executed in the apollo directory.

• descriptor_set_out is the descriptor file name
• modules/drivers/proto/sensor_image.proto the message define proto file

protoc --include_imports --descriptor_set_out=tmp modules/drivers/proto/sensor_image.proto

or you can use absolute path.

• descriptor_set_out is the descriptor file name
• proto_path the apollo home path
• /home/zero/01opencode/apollo/modules/drivers/proto/sensor_image.proto proto file absolute path

protoc --include_imports --descriptor_set_out=tmp --proto_path=/home/zero/01opencode/apollo /home/zero/01opencode/apollo/modules/drivers/proto/sensor_image.proto

2. Recover the record file.

• broken.record is the file need repair
• /apollo/sensor/camera/front_6mm/image the topic of the need repair message
• tmp the descriptor file generated in the previous step
• apollo.drivers.Image the message type of the need repair message

cyber_record recover -f broken.record -t /apollo/sensor/camera/front_6mm/image -d tmp -m apollo.drivers.Image

Examples

Below are some examples to help you read and write messages from record files.

1. Read messages

You can read messages directly from the record file in the following ways. If you need to speed up reading, you can use the Filter Read below.

from cyber_record.record import Record

file_name = "20210521122747.record.00000"
record = Record(file_name)
for topic, message, t in record.read_messages():
  print("{}, {}, {}".format(topic, type(message), t))

The following is the output log of the program

/apollo/localization/pose, <class 'LocalizationEstimate'>, 1627031535246897752
/apollo/canbus/chassis, <class 'Chassis'>, 1627031535246913234
/apollo/canbus/chassis, <class 'Chassis'>, 1627031535253680838

Filter Read

You can also read messages filtered by topics and time. This will improve the speed of parsing messages.

def read_filter_by_both():
  record = Record(file_name)
  for topic, message, t in record.read_messages('/apollo/canbus/chassis', \
      start_time=1627031535164278940, end_time=1627031535215164773):
    print("{}, {}, {}".format(topic, type(message), t))

2. Parse messages

To avoid introducing too many dependencies, you can save messages by record_msg.

pip3 install record_msg -U

record_msg provides 3 types of interfaces

csv format

you can use to_csv to format objects so that they can be easily saved in csv format.

f = open("message.csv", 'w')
writer = csv.writer(f)

def parse_pose(pose):
  '''
  save pose to csv file
  '''
  line = to_csv([pose.header.timestamp_sec, pose.pose])
  writer.writerow(line)

f.close()

image

you can use ImageParser to parse and save images.

image_parser = ImageParser(output_path='../test')
for topic, message, t in record.read_messages():
  if topic == "/apollo/sensor/camera/front_6mm/image":
    image_parser.parse(message)
    # or use timestamp as image file name
    # image_parser.parse(image, t)

lidar

you can use PointCloudParser to parse and save pointclouds.

pointcloud_parser = PointCloudParser('../test')
for topic, message, t in record.read_messages():
  if topic == "/apollo/sensor/lidar32/compensator/PointCloud2":
    pointcloud_parser.parse(message)
    # other modes, default is 'ascii'
    # pointcloud_parser.parse(message, mode='binary')
    # pointcloud_parser.parse(message, mode='binary_compressed')

3. Write messages

You can now also build record by messages. You can write pb_message by record.write.

def write_message():
  pb_map = map_pb2.Map()
  pb_map.header.version = 'hello'.encode()

  with Record(write_file_name, mode='w') as record:
    record.write('/apollo/map', pb_map, int(time.time() * 1e9))

Its application scenario is to convert dataset into record files. Please note that it must be written in chronological order.

If you want to write raw message, you should first use Builder to help convert raw data to pb_message.

image

You can write image to record file like below. ImageBuilder will help you convert image to pb_image. encoding should be rgb8,bgr8 or gray, y.

def write_image():
  image_builder = ImageBuilder()
  write_file_name = "example_w.record.00002"
  with Record(write_file_name, mode='w') as record:
    img_path = 'test.jpg'
    pb_image = image_builder.build(img_path, encoding='rgb8')
    record.write('/apollo/sensor/camera/front_6mm/image',
                 pb_image,
                 int(time.time() * 1e9))

lidar

You can write image to record file like below. PointCloudBuilder will help you convert pcd file to pb_point_cloud.

def write_point_cloud():
  point_cloud_builder = PointCloudBuilder()
  write_file_name = "example_w.record.00003"
  with Record(write_file_name, mode='w') as record:
    pcd_path = 'test.pcd'
    pb_point_cloud = point_cloud_builder.build(pcd_path)
    record.write('/apollo/sensor/lidar32/compensator/PointCloud2',
                 pb_point_cloud,
                 int(time.time() * 1e9))