Download the Waymo Open Dataset

Motion Dataset

• [new] v1.3.0 October 2024: Improved traffic light time stamp alignment. Note the current leaderboards for Motion Prediction are based on v1.2, and a new leaderboard for v1.3.0 will be updated in the upcoming months. - files

• v1.2.1 March 2024: Added camera embeddings data including 8 cameras with 360 degree coverage. Note this data provides embeddings only, not full images. - files

• v1.2, March 2023: Added Lidar data and driveway labels. Increased the max number of map points in tf_examples to 30k and reduced sampling to 1.0m to increase map coverage, so the coverage equalizes that of the dataset in scenario proto format. - files

• v1.2, March 2023: Added Lidar data and driveways. Increased the max number of map points in tf_examples to 30k and reduced sampling to 1.0m to increase map coverage, so the coverage equalizes that of the dataset in scenario proto format. - files

• v1.1, August 2021: Added lane connections, lane boundaries, and lane neighbors. See Github page for technical specs. - files

• v1.0, March 2021: Initial release - 103,354 segments with maps data - files

Perception Dataset

• [new]March 2024: we fixed several small errors in the 3D semantic segmentation ground truth labels, especially for the class of motorcyclist. - v1.4.3 files - v2.0.1 files

• v2.0.0, March 2023: Introduced the dataset in modular format, enabling users to selectively download only the components they need. - files

• v1.4.2, March 2023: Added 3D map data as polylines or polygons. For 2D panoptic segmentation labels, added a mask to indicate pixels covered by multiple cameras. - files

• v2.0.0, August 2023: Added the Perception object assets data - files

• v2.0.0, March 2023: Introduced the dataset in modular format, enabling users to selectively download only the components they need. - file link same as abovev1.4.2, March 2023: Added 3D map data as polylines or polygons. For 2D panoptic segmentation labels, added a mask to indicate pixels covered by multiple cameras. - files

• v1.4.1, December 2022: Improved the quality of the 2D video panoptic segmentation labels - files

• v1.4.0, June 2022: Added 2D video panoptic segmentation labels - files

• v1.3.2, May 2022: Improved the quality of the 3D semantic segmentation labels and 2D key point labels. Added a new field for the 3D Camera-Only Detection Challenge. - files

• v1.3.1, April 2022: Updated the dataset for the 3D Camera-Only Detection Challenge. This version includes a new test set with 80 segments, and additional fields for training and validation - tar files, individual files

• v1.3.0, March 2022: Added 3D semantic segmentation labels, 2D and 3D key points labels, and 2D-to-3D label correspondence labels - files

• v1.2, March 2020: Added Test Set with 150 segments, plus 800 segments for domain adaptation across Training, Validation, and Test - tar files, individual files

• v1.1, February 2020: Added camera labels for 900 segments - tar files, individual files

• v1.0, August 2019: Initial release - tar files, individual files

Community Contributions

OmniNOCS Dataset

• OmniNOCS Waymo Dataset Files

• Released in 2024 by Google Research

• OmniNOCS is a unified NOCS (Normalized Object Coordinate Space) dataset that contains data across different domains with 90+ object classes. The dataset includes NOCS coordinates, object instance masks and 3D bounding box annotations. The link above points to data based on the Waymo Open Dataset. For additional datasets please see the project page below.

• GitHub

WOMD-Reasoning Dataset

• WOMD-Reasoning Dataset files

• Released in 2024 by University of California, Berkeley

• WOMD-Reasoning is a language annotation dataset built on the Waymo Open Motion Dataset, with a focus on describing and reasoning interactions and intentions in driving scenarios. It presents by far the largest Q&A dataset on real-world driving scenarios, with around 3 million Q&As covering various topics of autonomous driving from map descriptions, motion status descriptions, to narratives and analyses of agents’ interactions, behaviors, and intentions.

• Please note that the languages contained in the current version are auto-labeled, and there may be subsequent release with human verification and improved accuracy.

• GitHub, and paper: "WOMD-Reasoning: A Large-Scale Language Dataset for Interaction and Driving Intentions Reasoning"

ROAD-Waymo Dataset

• ROAD-Waymo Dataset files

• Released in 2023, by Oxford Brookes University

• The ROAD-Waymo dataset builds on the Waymo Open Dataset, by adding the following annotations: action labels (turning-right, moving-away, etc.), agent type labels (pedestrians, large-vehicle, etc.), semantic location labels (in vehicle lane, in right pavement, etc.). The dataset is used in the Second Workshop & Challenge on Event Detection for Situation Awareness in Autonomous Driving at ICCV'23.

• GitHub

Causal Agents Labels

• Causal Agent Labels files

• Released in 2022, by Google Brain and Waymo

• Based on the Waymo Open Motion Dataset, the causal agent labels identify agents whose presence influences human drivers' behavior in any format. In addition to causal agent labels, the team also release perturbed copies of the Waymo Open Motion Dataset validation dataset, which serve as robustness benchmarks to aid the research community in building more reliable and safe models for motion forecasting.

• GitHub, and paper: "CausalAgents: A Robustness Benchmark for Motion Forecasting using Causal Relationships"

Scene Flow Labels

• Scene flow labels files

• Released in 2021, by Google Brain and Waymo

• The scene flow labels builds on Perception v1.2 of the Waymo Open Dataset. This release comprises supervised labels for each point cloud reflection with a direction and magnitude of motion at each time step. These labels are derived from tracked bounding boxes associated with labeled objects in the scene.

• Paper: "Scalable Scene Flow from Point Clouds in the Real World"