ChangeIn

ChangeIn is the first real-world benchmark that provides per-frame ground truth camera intrinsics for videos with dynamic intrinsics, and current baselines struggle to predict accurate intrinsics on our benchmark. This is the code repository for it.

Installation

To install dependencies for LUT generation and synthetic experiments, please install infinigen and then run ./install.sh.

To install dependencies used for evaluation, run evaluation/install_eval.sh.

Installation assumes the use of mamba, but you can replace instances of this within install scripts with conda for anaconda support.

Repository Organization

2. asd

3. Real Drone Experiments: QGroundControl

   Code relevant to flying the drone programmatically can be found under drone-qgc. Use create_drone_mission.py to create .plan missions to fly.

4. Real Drone Experiments: RPi Interface

   To ensure 2D and 3D point collection via drone calibration occurred at the same time, we placed a Raspberry Pi (RPi) onboard our drone and wrote a program that syncs turning on LED light with recording 3D RTK position. Files relevant to this can be found under drone-rtk, and documentation for setup can be found under drone-rtk/documentation_rtk.

5. Synthetic Board Experiments

   Synthetic board code can be found in the boards folder. Board designs can be found under boards/board_designs. All board experiments can be run via python run_all_boards.py, assuming that Kalibr docker has been built.

6. Synthetic Drone Experiments

   Synthetic drone code can be found in the drones folder. All drone experiments can be run via python run_all_drones.py, assuming that Kalibr docker has been built.

7. Running Baseline Methods

   Baseline method codebases can be found in the baseline_methods folder. Inference on image folders can be run via

   python universal_evaluation_script.py --method_name {colmap, droidcalib, geocalib} --image_folder {PATH TO IMAGES} --eval_intrinsics --time