Skip to content

Latest commit

 

History

History
51 lines (41 loc) · 1.83 KB

read_label.md

File metadata and controls

51 lines (41 loc) · 1.83 KB

How to read labeled data

Label Folder Structure

.
├── (rosbag_name)
    ├── img                       # Folder of camera images
        ├── img0000.png           # Camera Image 0
        ├── img0001.png           # Camera Image 1...
      ....
    ├── labels                    # Folder of label Txt files
        ├──bev0000.txt            # Label for Frame 1
        ├──bev0001.txt            # Label for Frame 2
      ...
    ├── velo                      # Birds'Eye View Image of Lidar
        ├──bev0000.png            # Lidar BEV frame 1
        ├──bev0001.png            # Lidar BEV frame 2
        ├──bev0002.png
      ...
        ├──velotimestamps.csv     # Timestamp and sequences of all BEV frames

    ├── (rosbag_name).bag         # Actual Rosbag
├── plot_GT.py                    # Utility Script to plot ground truth labels
├── via2label.py                  # Utility Script to generate label txt files from VIA's ugly json file
.

Labeling tools Used

VGG Image Annotator (VIA) from Visual Geometry Group in University of Oxford [Link] (https://www.robots.ox.ac.uk/~vgg/software/via/)

Label Format

All labels are given in the Velodyne frame and in metric space

An object entry contains the Following:

Name, width , height centroidx, centroidy, yaw

where yaw is measured from x axis and positive if the car is tilted to the left.

Example:

ADD EXAMPLE

Coordinate Frame Used

x-axis [0, 30] represent the distance in front of the car [front is positive, back is negative] y-axis [-20, 20] represent the distance beside the car. [left is positive, right is negative] z-axis [up is positive, bottom is negative]

The system is right-handed