A vehicle detection database for vision tasks set in the real world.

3D Reconstructions

Each photograph in NYC3DCars has been geo-registered to the Earth, providing full camera intrinsics and extrinsics in an Earth-Centered, Earth-Fixed coordinate system enabling seamless integration with existing geospatial data.

Geographic Data

Companion databases such as those provided by OpenStreetMap and NYC OpenData have been integrated for easy access to geographic features such as road, sidewalk, and median polygons as well as road network connectivity.

Vehicle Annotations

Human annotators have provided detailed descriptions for vehicles contained in the database. Annotations include a full 6 degree of freedom vehicle pose, vehicle type, 2D vehicle bounding box, and approximate photo time of day.


Use of PostgreSQL with the PostGIS package is recommended for working with any of the spatial or geographic data and is required for the use of many of the analysis tools. JPEG images are distributed separately per reconstruction due to the large size. We provide a link below to our preprocessed geodatabase as well as links to original sources.

Files labeled as PGSQL can be imported into a PostgreSQL database using the following command:

zcat filename.dump.gz > | pg_restore --dbname=nyc3dcars

Alternatively, we provide a script to help download and import all necessary assets.

To enable the reproduction of our detection results, we have included the deformable part models that we trained using a combination of our dataset, the PASCAL Visual Object Classes Challenge 2007 dataset, and the KITTI Vision Benchmark Suite dataset. In addition, we provide links to our open source detection software, pydro (a python reimplementation of Pedro Felzenszwalb and Ross Girshick's voc-release5 MATLAB package), our nyc3dcars-toolkit analysis scripts, and the source code for our annotation website.

Creative Commons License
NYC3DCars by Kevin Matzen is licensed under a Creative Commons Attribution 3.0 Unported License.


NYC3DCars: A Dataset of 3D Vehicles in Geographic Context

Kevin Matzen, Noah Snavely

To appear in Proc ICCV 2013

If you wish to use this dataset or any of the related code for use in your own publication, please include the citation listed below.

    author = {Matzen, Kevin and Snavely, Noah},
    title = {NYC3DCars: A Dataset of 3D Vehicles in Geographic Context},
    booktitle = {Proc. Int. Conf. on Computer Vision},
    year = {2013},