Calibench: Community Datasets

Community-contributed calibration datasets for the Calibench benchmark — sensor rigs, lenses, and capture conditions beyond the official collection.

Contributing a dataset

Open a PR to this repo (Community tab → New pull request) with a folder per config (<your-rig>/ holding the sequences + target config) and a <your-rig>/README.md filled from TEMPLATE/README.md. Filled examples (our own MVIS rig): virig-3cams · virig-4cams. The metadata below is exactly one row of the official dataset catalog — declare it so your dataset drops into the catalog (and becomes eligible as a leaderboard roster) without guesswork.

These requirements are themselves open to improvement — PR this README or use the Community tab to propose changes.

Required metadata (✅ required · ◽ recommended)

Field	Req.	What to provide	Example
Sensors — per camera
Camera sensor	✅	Sensor model / part number	OmniVision OV9282
Resolution / fps	✅	Per camera	848×800 @ 30 Hz
Shutter	✅	global / rolling; if rolling: line readout time if known	rolling, 29.47 µs/row
Color / encoding	✅	mono or color; pixel encoding	mono8
Lens	◽	Lens model + focal length if known	Lensagon BF2M2020S23
FOV	✅	Degrees (state diagonal or H×V); fisheye/wide/standard	195° diagonal, fisheye
Sensors — per IMU (if any)
IMU model	✅	Part number	Bosch BMI160
Rates	✅	Gyro and accel rates separately if they differ	gyro 200 Hz, accel 62.5 Hz
Noise characteristics	◽	Noise density / bias instability if characterized	from Allan variance
Rig / device
Synchronization	✅	hw-trigger / sw / free-running clocks; known offsets; which clock stamps the data	FPGA-synced cams+IMU
Rig layout	✅	#cams/#IMUs, arrangement, nominal baselines, rigid mounting	stereo, 11 cm baseline
Rig photo	◽	Strongly encouraged	rig.jpg
Datasheet links	◽	Manufacturer page/PDF per sensor	intel.com/…/t265
Reference calibration	◽	Factory/prior calibration with provenance (who, which tool, when) — reference, not ground truth unless it has uncertainty	T265 factory intrinsics
Calibration target
Target type + geometry	✅	aprilgrid/checkerboard/ChArUco; rows×cols, tag/square size (measured, not nominal — say which), spacing, family	AprilGrid 6×6, 8.8 cm, t36h11
Target config file	✅	The file itself, in the dataset folder	april.yaml
Physical realization	◽	Material/flatness; how the size was verified	foam board, calipers
Capture & data
Sequences	✅	Count + what distinguishes them; same physical rig per config? (sequences become repeatability folds)	5 bags, same rig
Capture protocol	✅	Motion (axes excited), duration, lighting, known artifacts (blur, drops)	full 6-DoF excitation, indoor
Format	✅	ROS bag preferred (one per sequence) + topic list; else timestamped folders + IMU CSV with exact format	/cam0/image_raw …
Ground truth	◽	Mocap/laser-tracker: system, rate, accuracy	OptiTrack @ 120 Hz
Legal
License	✅	CC-BY-4.0 grant for hosted data, or a download.py recipe (manifest + sha256) for data you cannot relicense — never re-hosted third-party bytes	CC-BY-4.0
Citation	✅	Required BibTeX	@article{…}
Contact	✅	Maintainer contact	you@lab.edu

By contributing you affirm you have the right to share the data under the license you grant. Accepted datasets are added to the Calibench collection and become eligible as leaderboard tracks/rosters.

Submitting results instead? See yangyulin/calibench-submissions (SUBMISSION.md).