This is the documentation on how to use the annotation tool provided by K-Lane
- Install ROS as directed by their site (we use ROS-Melodic)
- Create a new conda environment
conda create -n #env-name python=3.7
- Install the dependencies
pip install PyQt5 pyaml rospkg numpy tensorflow==1.15 opencv-python-headless
- note: installing opencv-python with PyQt5 may result in xcb error
KLaneDet
├── annot_tool
├── build
├── devel
├── frontal_image
├── point_cloud
├── src
├── gui_qt
├── temp
├── seq_1
├── bev_image
├── bev_image_label
├── bev_tensor
├── bev_tensor_label
├── lidar_msgs
├── pc_pre_processor
├── baseline
├── configs
├── data
├── logs
Before starting the labelling process, make sure that:
- A rosbag file containing recordings of both camera image and LiDAR point cloud is available
- The
/frontal_image/
,/point_cloud/
,/src/gui_qt/temp/
, and their children directories have been created
- Go to the
/annot_tool/
directory and setup the project by running
catkin_make
we need to run this line every time a change is introduced to the .cpp files
- Source the project's setup.bash, run the roscore, and play your rosbag file
source devel/setup.bash
roscore
rosbag play #path_to_your_rosbag_file
- Start the pointcloud preprocessor node
rosrun pc_pre_processor pc_pre_processor_node
- Start the annotation tool GUI
python /src/gui_qt/mainframe.py
The initial GUI will look like the image below
- Initialize the annotation tool and start annotating
- Make sure to initialize first for every new point cloud frame
- Tips: increasing the brightness of the display will help the annotation process significantly
- Do the post-processing and save the annotation
- The annotations and processed files can be found either on
/frontal_image/
,/point_cloud/
, or/src/gui_qt/temp/
under the sequence directory
- The annotations and processed files can be found either on