DreamVu launches 16K 360-degree real-world dataset for world models

DreamVu has released Liberty3D, a sample corpus of real-world scenes captured in 16K 360° color with measured metric depth and dense point clouds. The dataset is aimed at teams training world models and generative video models that need direct observations of appearance and geometry from the same instant.

Why it matters: - World models learn from observations of the physical world, and Liberty3D is designed to give them color and measured depth from the same capture instead of predicted geometry or reconstructed scans. - The dataset targets teams training world models, generative video models, robot systems, and simulation environments that need real-world 3D data. - High-resolution, metric depth data can help models learn the relationship between appearance and geometry more directly.

What happened: - DreamVu released Liberty3D, a sample corpus of landmarks across Philadelphia, New Jersey, and Washington, DC. - Each scene was captured with DreamVu’s Alia camera in a single shot from a single camera position. - The corpus is available at world model data for teams building world models and generative video models.

The details: - Liberty3D includes 16K color, metric depth at every pixel, and dense point clouds. - The captures cover the full 360° view. - DreamVu said the depth was measured at capture rather than inferred later. - The company said the point clouds can also serve as a starting point for Gaussian splats and other reconstructions. - Selected scenes can be explored in the browser with color, depth, and 3D point cloud views. - Access to the full corpus is available on request. - DreamVu runs custom capture programs for customer environments. - Alia is a single camera that captures 360° stereo in one shot. - DreamVu said the depth is aligned to the color image because both come from the same optics. - The optical design is patented and has grants in multiple countries. - The single-camera panoramic stereo method was published at CVPR. - DreamVu’s processing pipeline converts each capture into color, depth, and point cloud outputs without manual intervention.

Between the lines: - The launch positions DreamVu against two common alternatives: 360° consumer cameras that lack measured depth, and scan-based reconstruction pipelines that produce meshes or Gaussian splats after the fact. - DreamVu is framing Liberty3D as observation-first data, not a reconstruction, which is a meaningful distinction for teams trying to train models on raw world structure. - The company is also signaling that the same data may serve both model training and downstream 3D reconstruction workflows. - Sashi Reddi, Co-Founder and CEO of DreamVu, said the company built the camera and the pipeline, and invited teams to inspect scenes in 3D themselves.

What's next: - DreamVu said teams can request access to the full corpus. - The company said it will run capture programs in the environments customers need their models to learn from. - DreamVu will likely use Liberty3D to expand its pitch for custom data collection around specific training needs.

The bottom line: - Liberty3D is a high-resolution 360° dataset built to give AI models real-world color and metric depth from the same instant, which DreamVu says makes it better suited for world-model training than video alone or reconstructed 3D data.