Skip to content

Repository files navigation

Paris taxi dashcam dataset

Frames Route Capture window Device Collection mode Faces redacted Plates redacted
1,000 17.33 km 16:41:48.907Z to 18:18:51.835Z Xiaomi Redmi A5 Taxi dashcam 3,930 2,083

Paris Taxi Dashcam Dataset

1,000 consecutive, geotagged street images captured from a taxi in Paris on July 16, 2026.

The sequence runs from 16:41:48.907 to 18:18:51.835 UTC, or 18:41:48.907 to 20:18:51.835 CEST. A forward-facing Xiaomi Redmi A5 was fixed inside the taxi and used as a dashcam. Every delivered image is 1920 × 1080 pixels. GPS is available for all 1,000 frames.

Route

Route covered by the Paris taxi sequence

The route above shows the complete recorded taxi journey.

Sample frames

Sample frame showing a residential Paris street Sample frame showing a tree-lined Paris street Sample frame showing traffic on a Paris road
Traffic waiting at a red light on a broad Paris avenue. A cinema façade with a large billboard, entrance steps, and pedestrians. A narrow Paris street lined with shops and apartment buildings, with cyclists and pedestrians.

Dataset

Field Value
Images 1,000 consecutive JPEG frames
Resolution 1920 × 1080
Location Paris, France
Collection date July 16, 2026
Capture time 16:41:48.907Z to 18:18:51.835Z
Duration 1 hour, 37 minutes, 2.928 seconds
Route length Approximately 17.33 km, calculated from consecutive GPS points
Collection mode Forward-facing taxi dashcam
Device Xiaomi Redmi A5
GPS coverage 1,000 of 1,000 frames
Privacy processing 3,930 face regions and 2,083 license plate regions redacted
Package paris-eval-sample.zip, 422,779,239 bytes, approximately 403.2 MiB

Download

Download the Paris sample

1,000 privacy-redacted frames · GPS metadata · 403.2 MiB ZIP

Download

Downloading or using the dataset is subject to the GDC Dataset Evaluation License. For commercial licensing, production use, or a custom collection, contact ask@generaldatacorp.com.

Package structure

paris-eval-sample.zip
└── selected-sequence/
    ├── camera.json
    ├── manifest.csv
    └── images/
        ├── frame_000001.jpg
        ├── ...
        └── frame_001000.jpg

The package includes no raw, unredacted frames.

Repository files

File Purpose
manifest.csv One verified record per delivered image, including time, GPS, dimensions, status, filename, and checksum
route.geojson Route geometry generated from the 1,000 manifest coordinates
camera.json Device, mount, capture, positioning, privacy, and package metadata
DATASET_CARD.md Provenance, intended uses, privacy handling, and limitations
SCHEMA.md Exact field definitions and validation rules
checksums.sha256 SHA-256 checksum for the downloadable ZIP

Verified integrity

The archive opens successfully. It contains exactly 1,000 JPEG files and 1,000 manifest records. Every image filename exists, every sequence index is present, and all 1,000 image SHA-256 checksums match the manifest.

Privacy

Faces and license plates were automatically redacted before packaging. The export reports 3,930 face regions and 2,083 license plate regions redacted. Unredacted originals are retained separately under restricted access and are not included in this release.

Do not use the dataset to identify people or attempt to reverse privacy protections.

Scope

This is one continuous sequence from one taxi, one phone, one forward-facing viewpoint, and one evening in Paris. It is not representative of every district, season, weather condition, road type, or time of day. The release does not include LiDAR, stereo imagery, audio, video, 3D labels, or camera calibration.

License

The repository materials and downloadable dataset are provided under the GDC Dataset Evaluation License. The license permits limited internal evaluation. It does not permit redistribution, resale, sublicensing, public hosting, commercial use, or model training without written permission.

Copyright © 2026 Noir Technologies, Inc., doing business as The General Data Corporation. All rights reserved.