Data Visualization

Processed Scenes

To visualize the data you may have to install the following dependencies.

• Open3d
• PIL
• psutil
• matplotlib
• empy
• catkin_pkg
• rospkg

To use the prediction function, you should have ROS installed.

Introduction of Scripts

• ConcatLiDAR.py

  • Given a source directory, loads point clouds and transforms by their pose to world coordinates.
  • Iterates through frames, adding point clouds without removing previous ones.
  • The result is a scene with occlusions and traces, however it can be used to test the point clouds and coordinate transformations.

• VisualizeScene.py

  • Visualize completed semantic scene in either Cylindrical or Cartesian coordinates.
  • Samples points from each valid cell, and displays a colored point cloud in Open3D.
  • Also loads a bird's eye view image for each frame to verify the scene is correct.

• VisualizeDBKI.py

  • Similar to VisualizeScene.py, however visualizes semantic mapping data, which includes an extra split.
  • Not directly used for MotionSC.

Raw Data

• VisualizeLidar.py

  • Visualize raw sensor data before processed into a semantic scene.
  • Transforms all sensors to world coordinates, then displays the union of all semantic point clouds in Open3D.
  • Used to check all sensors.

Predictions

• plot_voxels.py

  • ROS Node for visualizing all baseline model predictions simultaneously with ground truth bird's eye view image.