extract_rosbag contains scripts to extract files from rosbags. ROS topics are extracted as image, mat or csv files.
To extract the files, go to the folder and run:
python extract_files.py
Data storage paths can be configured in 'config.yaml'. Platform can be selected in extract_files.py file itself.
radar_preprocess folder contains scripts to process radar data for milliEgo training.
The scripts do the following things:
- Overlay radar data. Overlays 3 adjacent radar frames (2 frames in the case of UAV) into 1 frame, in order to increase density and reduce fluctuation.
- Stitching 3 radars on UGV platform.
- Save radar data in the form of depth images
To pre-process the radar data from each platform, run:
python process_radar_handheld_uav.py
Or:
python process_radar_handheld.py
Platform can be selected by configuring inside of process_radar_handheld_uav.py.
We package sequences into h5 files in order to facilitate training.
create_dataset_milliego creates training dataset for milliEgo on Handheld and UAV platforms, where there's only 1 radar.
create_dataset_milliego_ugv creates training dataset for milliEgo on UGV platform. It uses data from 3 radars.
create_dataset_deeptio creates dataset for DeepTIO on Handheld and UGV platforms, where thermal images are 16bit.
create_dataset_deeptio_uav creates dataset for DeepTIO on UAV platform, where thermal images are 8bit, and RGB images need to be reshaped.
The scripts do the following things:
- Normalize image data
- Align different modalities by associating timestamps
- Package data into h5 files
Configure the parameters in config.yaml.
To create the datasets, simply go to corresponding folder, and run:
python os_create_dataset.py