Visualized using Foxglove: Wayve's WayveScenes101 Dataset
Wayve's WayveScenes101 is a groundbreaking dataset tailored to propel advancements in novel view synthesis for autonomous driving. This dataset features 101 diverse driving scenarios, encompassing urban, suburban, and highway environments, all captured under varying weather and lighting conditions. Each 20-second scene includes time-synchronized imagery from five vehicle-mounted cameras, delivering an impressive 101,000 high-resolution images at 10 frames per second. Enriched with COLMAP-derived camera poses and 3D points, WayveScenes101 ensures precision in scene reconstruction.
Notably, the dataset features a held-out evaluation camera, enabling off-axis reconstruction tests to evaluate model generalization. Detailed metadata for each scene enhances nuanced performance evaluations across a range of conditions. Seamlessly integrating with the NerfStudio framework, WayveScenes101 provides a robust foundation for benchmarking and refining view synthesis models in dynamic driving contexts.
Complementing this dataset are Wayve’s cutting-edge AI advancements:
1. GAIA (Generative AI for Autonomy): A generative world model capable of producing realistic driving videos from text and video inputs. GAIA accurately interprets prompts to generate diverse traffic scenarios, motions, time-of-day and weather conditions, and realistic interactions between road users.
2. LINGO (Language INterfaced Generative mOdel): This model integrates natural language into autonomous driving AI, enhancing explainability and trustworthiness. By incorporating causal reasoning and descriptions of driving actions, LINGO accelerates learning for new and rare scenarios, while offering deeper insight into scene understanding, reasoning, and decision-making.
Together, WayveScenes101, GAIA, and LINGO represent major strides in autonomous driving research. Thanks for the awesome dataset and research, Wayve! 🙏