Dataset Gallery

This page is the adoption map for public and internal datasets. A dataset is useful to libgnss++ when it has a repeatable command, a clear reference source, and an artifact contract.

Current lanes

Dataset	Status	Best entrypoint	Reference	Main use
PPC-Dataset Tokyo/Nagoya	bundled primary RTK sign-off	gnss ppc-rtk-signoff	reference.csv trajectory truth	Moving RTK coverage and fix quality
UrbanNav Tokyo Odaiba/Shinjuku	path-override public smoke	gnss ppc-rtk-signoff --run-dir ...	Applanix reference export	Urban canyon RTK stress
smartLoc urban GNSS	receiver-fix lane	gnss smartloc-adapter, gnss smartloc-signoff	NovAtel SPAN differential RTK/IMU	NLOS receiver-fix analysis
Google Smartphone Decimeter Challenge	candidate	new adapter needed	competition ground truth	Smartphone raw GNSS and IMU stress
Ford Highway Driving RTK	candidate	new adapter needed	INS plus survey-grade GNSS	Highway-scale coverage
Oxford RobotCar RTK ground truth	candidate	new adapter needed	post-processed GPS/IMU reference	Long-term localization reference

Use:

python3 apps/gnss.py public-rtk-benchmarks --format markdown

to print the current public benchmark inventory and caveats.

Bundled PPC-Dataset

The repo includes a PPC-Dataset checkout under data/PPC-Dataset. Each run has the same structure:

data/PPC-Dataset/<city>/<run>/
  rover.obs
  base.obs
  base.nav
  reference.csv
  imu.csv
  trajectory.kml

Run a short smoke:

python3 apps/gnss.py ppc-rtk-signoff \
  --dataset-root data/PPC-Dataset \
  --city nagoya \
  --run run1 \
  --max-epochs 200 \
  --summary-json output/adoption/ppc_nagoya_run1_smoke.json

Run the six-run matrix when evaluating a profile:

python3 apps/gnss.py ppc-coverage-matrix \
  --dataset-root data/PPC-Dataset \
  --summary-json output/adoption/ppc_coverage_matrix.json

External RTK dataset pattern

When a dataset can be mapped to rover/base/nav/reference files, keep the adapter thin and run the normal sign-off path:

python3 apps/gnss.py ppc-rtk-signoff \
  --run-dir /datasets/my_rtk_dataset/run1 \
  --rover /datasets/my_rtk_dataset/run1/rover.obs \
  --base /datasets/my_rtk_dataset/run1/base.obs \
  --nav /datasets/my_rtk_dataset/run1/base.nav \
  --reference-csv /datasets/my_rtk_dataset/run1/reference.csv \
  --out output/my_rtk_dataset/run1.pos \
  --summary-json output/my_rtk_dataset/run1.json

Only write a new adapter when the dataset layout, timestamp convention, or reference format cannot be expressed with path overrides.

Dataset adapter contract

A new adapter should produce a normalized run directory with:

File	Required	Notes
rover.obs	yes	RINEX observation file for the rover
base.obs	RTK only	RINEX observation file for the base station
base.nav or navigation.nav	yes	Broadcast navigation file
reference.csv	sign-off runs	Time-aligned truth or reference trajectory
receiver_solution.csv	optional	Commercial or receiver-engine comparison
provenance.json	recommended	Receiver, antenna, source URL, conversion command

The adapter should also document:

• source dataset version or download date,
• coordinate frame and time scale,
• antenna and receiver metadata if available,
• any filtering or interpolation applied to reference truth,
• the exact command that produces the normalized directory.

Promotion rule

Treat a dataset lane as stable only when it has:

1. one smoke command with --max-epochs,
2. one full sign-off command,
3. at least one checked summary.json,
4. docs that explain the reference source and caveats.

This keeps public claims tied to commands instead of screenshots or ad hoc logs.