The Berkeley DeepDrive 100K Dataset for Advanced Autonomous Driving Research

The Berkeley DeepDrive 100K, also known as the BDD 100K, is a comprehensive dataset tailored for the autonomous driving community. It offers a substantial volume of data essential for the development and refinement of computer vision models.

BDD100K is a dataset of autonomous driving by the Berkeley Artificial Intelligence Research Lab in the USA for the transportation and mobility sector.

Capturing a wide array of driving conditions, the BDD 100K dataset stands out as a significant resource for computer vision engineers and data scientists.

Data collection

The BDD 100K dataset distinguishes itself with its impressive collection of over 100,000 video sequences from more than 50K rides. Each video is 40 seconds long at 30 frames per second.

These videos are recorded under diverse geographic and environmental conditions, offering an extensive range of data crucial for training and testing computer vision systems.

Real-World Scenarios

The dataset encompasses a broad spectrum of driving scenarios, providing a realistic and challenging environment for testing and improving autonomous driving algorithms.


The BDD100K does not include the multi-modal data found in some of its more recent counterparts. It is still one of the most extensive dataset, featuring 120 million image frames.

Multifaceted Annotations

Each video in the BDD 100K dataset comes with detailed annotations, including lane markings, traffic signs, and object tracking. This level of detail is instrumental in developing a nuanced understanding of driving scenarios and ensures precise model training.

Preview and Usage

You can preview and copy a sample of the BDD 100K dataset on, offering a glimpse into the dataset’s capabilities and applications. The link for previewing is here:

Diving into the BDD 100K Dataset

The BDD 100K dataset serves as a key resource for advancing research in autonomous driving. Its extensive coverage of different driving conditions, combined with detailed annotations, makes it a valuable tool for developing robust and accurate computer vision models.

Access to the BDD 100K dataset requires an account, but the process is straightforward, ensuring secure and efficient download of the data.

Dataset information

Share this dataset