Tool for labeling laser scan data from ROS bags
On Ubuntu 16.04 (should work on other ubuntu versions, just replace 'kinetic' with 'indigo' etc.:
- Install ROS Kinetic
sudo apt-get install python-qt4 ros-kinetic-std-msgs python-numpy python-scipy
To get started run:
python src/qt_labeler.py
Then open a bag file containing laser scans.
- Click empty space to create a new label region
- Click and drag a region to move it
- Clicking a region start or end moves the start/end to the new location
- Clicking a region middle (blue or green) split the region at the current frame
- Regions are interpolated linearly between start and end pose
- Using the scroll wheel over a region enlarges or shrinks the region radius
- Middle clicking a region deletes the end of the region (making it extend to the end of the data)
- Right clicking a region deletes it (moving the end to the previous frame)
- Teal: Region start frame
- Blue: Region middle frame
- Green: Region goes to end of data
- Red: Region end frame
- Use the buttons to move forward/backward in time (or arrow keys left/right)
- Press space to play/pause the data
The output is in the form of a matlab/octave/scipy.io compatible mat file. To load it in python use scipy.io.loadmat.
Entries: N=number of scans, C=number of returns per scan
| key | type | description |
|---|---|---|
| range_max | float | the max range of the laser scan message |
| theta | float array[C] | the angular offset of each point |
| scans | float array[N,C] | the input data. X in typical ML problems |
| classes | float array[N,C] | the labeled data. Y in typical ML problems |
- The app automatically grabs the first laster scan topic it finds. To run it on other topics, filter out the topic you want using
rosbag filter - The app currently only offers binary labeling (0=default, 1=in a marked region)