- At first you need to preprocess the dataset. Preprocessing includes the transformation of depth images into RGB images coordinates and converting pictures to undistorted format. You can do this by executing:
python preprocessing.py PATH_TO_DATASET PATH_TO_SAVE_COLOR PATH_TO_SAVE_DEPTH INI_CONFIG_FILE
WARNING: You should have requirements installed for running script.
NOTE: You should specify preprocessing_config.ini file before running script. There is example of this file in the repository.
NOTE: You should have color and depth folders in dataset.
- Then you can generate your own associations file for one-view or two-view cases:
python associate_one_view.py RGB_DIR DEPTH_DIR > associations.txt
python associate_two_view.py RGB_MASTER_DIR DEPTH_MASTER_DIR RGB_SLAVE_DIR DEPTH_SLAVE_DIR > associations.txt
NOTE: You can specify max_difference and timestamp2sec parameters. By default they equals 1000 microseconds and 1e6 accordingly.
- Now you can run ORB-SLAM3. For one-view version use:
./Examples/RGB-D/rgbd_tum Vocabulary/ORBvoc.txt SETTINGS_YAML_FILE PATH_TO_SEQUENCE_FOLDER ASSOCIATIONS_FILE
For two-view:
./Examples/RGB-D-Two-View/rgbd_tum_tw Vocabulary/ORBvoc.txt SETTINGS_YAML_FILE PATH_TO_SEQUENCE_FOLDER ASSOCIATIONS_FILE
NOTE: The repository already has a yaml configuration file example. All Slave parameters and TransformationMatrix don't affect anything in the case of one-view version.