This repo contains code to process the RobotCar dataset.
Go into the repo directory on HPC /work/qvpr/workspace/RobotCar/
and run conda activate ./envs
to activate the conda environment with required dependencies to run the scripts.
Assuming your pwd is in the base directory of this repo, first use pip install -e .
to install the repo as a package.
In addition, go to src/settings.py
and change the RAW_DIR
variable to be the base directory for the raw RobotCar dataset on your local machine. Download the raw data from HPC at path /work/qvpr/data/raw/RobotCar/
using rsync or however you want.
To setup the conda environment will all dependencies, use conda create --name RobotCar --file requirements.txt
in the base directory of the repo.
This codebase allows you to undistort and deBayerize the raw images using the src/process_raw/ready_images.py
script and saves them to the path specified under READY_DIR
in src/settings.py
.
This codebase allows you to interpolate the RTK poses to the camera image timestamps for both stereo and monocular cameras. The camera extrinsics relative to the GPS is given, providing poses in a globally consistent coordinate frame for all images. See the src/process_raw/gps_camera_align.py
script more more information around usage. The output of the poses will be in the file $READY_DIR/traverse_name/camera_name/camera_poses.csv
in xyzrpy format.
The src/utils/geometry.py
file contains some useful functions and classes for representing and performing operations with 6DOF transformations that will come in handy for finding image correspondences.