Skip to content

Reproducing the PVS dataset in Telemachus format

The PVS dataset (Menegazzo & von Wangenheim, 2020 — Curitiba, Brazil) is licensed under CC-BY-NC-ND-4.0, which forbids redistribution of derivative works. As a consequence:

  • PVS is not on Zenodo in Telemachus format (unlike AEGIS, STRIDE, RS3).
  • You need to download the raw CSVs from Kaggle yourself, then convert locally with the bundled tele adapter.

If you just want to try Telemachus on a real IMU-rich dataset without going through Kaggle, reach for AEGIS or STRIDE instead — both ship ready-to-use parquets.

What PVS brings

Property Value
Hardware InvenSense MPU-9250, 2 sensors per vehicle (left / right)
Accel / Gyro / Magneto 100 Hz ground truth
GPS ~1 Hz
Trips 9 (3 vehicles × 3 drivers × 3 sensor placements)
Rows 1,080,905
Location Curitiba, Paraná, Brazil — Dec 2019
License CC-BY-NC-ND-4.0

Strong points: 3 placements per trip (dashboard / above-suspension / below-suspension), road surface labels, dual-sensor redundancy. Weak point: can't be republished.

Reproduction — 3 steps

The full step-by-step (with Kaggle CLI, expected tree, checksum verification) lives next to the converter in datasets/pvs/README.md. In short:

pip install telemachus kaggle

# 1. Download raw data from Kaggle
kaggle datasets download -d jefmenegazzo/pvs-passive-vehicular-sensors-datasets
unzip pvs-passive-vehicular-sensors-datasets.zip -d /path/to/pvs/

# 2. Convert to Telemachus (pick a placement + side)
tele convert pvs /path/to/pvs/ -o datasets/pvs/ --placement dashboard --side left

# 3. Validate
tele validate datasets/pvs/ --level basic

Citing PVS

When you publish results using PVS data, cite both the original dataset and the Telemachus format spec:

  • Menegazzo, J. & von Wangenheim, A. (2020). PVS — Passive Vehicular Sensors datasets. Kaggle.
  • Edet, S. (2026). Telemachus Specification v0.8. Zenodo. DOI 10.5281/zenodo.19609019.