Open Datasets

It's often difficult to find open source and accessible datasets, that reflect multi-sensor environments with annotations, to use for testing and trial. We're compiling an updated list for reference here - follow the links to get the dataset!

ZOD - Zenseact Open Dataset

The Zenseact Open Dataset (ZOD) is a large multi-modal autonomous driving (AD) dataset, created by our customer, Zenseact. Over a 2-year period in 14 different European counties, researchers at Zenseact collected data using a fleet of vehicles equipped with a full sensor suite consisting of 3x lidars, 1x camera and IMU data. In total, the dataset includes 100k single frame images, 1473 sequences and 29 Drives. 

Access ZOD here

Audi A2D2

Audi's dataset includes 40k frames with semantic segmentation labels, 12k with 3D bounding boxes and approximately 390k unlabeled frames in sequences from several German cities. It is fed by a multi-sensor matrix with 5x Lidars, and 6x cameras capturing a dense 360 view of the environment in 2D and 3D.

Access A2D2 here

Berkeley BDD100K

This open source dataset is very diverse, including data from different times of the day, different weather scenarios that were captured from urban, highway and rural areas.

Access BDD100K here


NuScenes is a multi-modal dataset with 6 cameras, lidar, IMU & GPS. It consists of 1000 dense urban scenes collected from test fleets that were active in both Boston and Singapore.

Access NuScenes here