The Boxy Vehicles Dataset

A large vehicle detection dataset with almost two million annotated vehicles
for training and evaluating object detection methods for self-driving cars on freeways..
Dataset Code --- Website Code

Boxy sample vehicle labels

Quick specs:

  • 200,000 images
  • 1,990,000 annotated vehicles
  • 5 Megapixel resolution
  • Sunshine, rain, dusk, night
  • Clear freeways, heavy traffic, traffic jams

Annotation with truck carrying a car


2D boxes or 3D cuboids for vehicles

Axis-aligned bounding boxes (AABB) and trapezoids for a more accurate description.
Simple AABB may intrude into neighboring lanes, as shown on the right.

Modern Vehicle Detection

Which method do you choose? What do you focus on?
High accuracy, low memory usage, low computational complexity, realtime detections, embedded systems,...

Your Research

Try our benchmarks, use the data to train a classifier, use it for domain transfer, or create new metrics.
We do encourage new benchmarks, let us know if additional data (e.g. odometry information) would be useful, and feel free to extend the dataset's scripts on Github.

Citation

The paper has been accepted and is undergoing final changes.
@inproceedings{boxy2019,
   title={Boxy Vehicle Detection in Large Images},
   author={Behrendt, Karsten},
   booktitle={Proceedings of the IEEE International Conference on Computer Vision},    year={2019}
}