Data Processing

Data processing tools from bag files to data sequences. The dataset is used to train MonoForce traversability estimation models.

The bag files are available at:

• http://subtdata.felk.cvut.cz/outdoor-dataset/
• http://subtdata.felk.cvut.cz/robingas/

Usage

Make sure to adjust the paths and data topics.

• To save lidar clouds, corresponding camera images, and calibration (extrinsics and intrinsics) from a bag file:

  OUTPUT_PATH=/path/to/save/data/sequence
  roslaunch dataproc dataproc.launch output_path:=${OUTPUT_PATH} img_topics:=[] lidar_topics:=[] camera_info_topics:=[]
  

• To save control inputs:

  cd ./scripts/
  ./add_cmd_vels
  

• RGB data anonymization using the Deface package:

  cd ./scripts/
  ./blur_faces.sh
  

• Save semantic pseudo labels using the SEEM model:

  cd ./scripts/
  ./save_semantic_pseudolabels.sh
  

• Save semantic pseudo labels using the WildScenes models:

  cd ./scripts/
  ./save_semantic_pseudolabels_wildscenes
  

  The script is based on the mmsegmentation inference tutorial. Please make sure to install the required dependencies and download the WildScenes pretrained models.

• Save localization (lidar poses). The norlab_icp_mapper SLAM was used to obtain the poses:

  cd ./scripts/
  ./add_lidar_poses