Waymo Releases Massive Open Dataset to Advance Self-Driving Research
Waymo launches the Waymo Open Dataset, offering researchers free access to high-resolution sensor data collected across 1,000 driving segments. This diverse dataset aims to accelerate machine learning innovations in autonomous vehicle perception and prediction.
Waymo introduces the Waymo Open Dataset, a free, high-quality multimodal sensor dataset designed for autonomous driving research. Collected from over 10 million autonomous miles driven in 25 cities, this release gives researchers access to high-resolution data gathered by Waymo's self-driving vehicles. The company views this massive data release as a way to turn academic ideas into real innovations by providing the critical ingredient needed for machine learning.
The dataset contains 1,000 driving segments that each capture 20 seconds of continuous driving, resulting in 200,000 frames at 10 Hz per sensor. It covers diverse environments across Phoenix, Kirkland, Mountain View, and San Francisco, featuring a wide spectrum of conditions like day, night, dawn, dusk, sunshine, and rain. Each segment includes synchronized data from five high-resolution lidars and five front-and-side-facing cameras, providing a complete 360-degree view of the road.
To make the data immediately useful for machine learning, Waymo includes dense labeling across the footage with 12 million 3D labels and 1.2 million 2D labels identifying vehicles, pedestrians, cyclists, and signage. This careful synchronization of camera and lidar data allows researchers to develop advanced 3D perception models that fuse multiple sensor inputs. Ultimately, Waymo hopes this rich and diverse collection accelerates global advances in 2D and 3D perception for the entire self-driving industry.