DreamVu launches Liberty3D, a corpus of real-world scenes in 16K 360° with metric depth, for teams training world models

A breakthrough dataset providing 16K 360° visual coverage with measured metric depth to train AI world models on direct real-world observations.

PHILADELPHIA, PA, UNITED STATES, September 15, 2026 /EINPresswire.com/ — DreamVu today released Liberty3D, a sample corpus of landmarks across Philadelphia, New Jersey, and Washington, DC, captured with the company’s Alia camera and delivered as 16K color, metric depth at every pixel, and dense point clouds. Every scene was captured in a single shot from a single camera position. The corpus is available to teams building world models and generative video models at dreamvu.ai/world-model-data.

World models learn the structure of the physical world from what they are shown. Most of what they are shown today is video without geometry, or geometry reconstructed after the fact. Liberty3D is built to give these models something different: the color and the metric depth of the same instant, from the same optics, at a resolution high enough to see the geometry of a real place directly.

Observations, not reconstructions
Teams today get real-world 3D data in two ways. Consumer 360° cameras such as Insta360 record high-quality color, but no measured depth. Depth has to be predicted afterward by a model, and a world model trained on predicted depth inherits the model’s errors. Handheld scanners such as XGRIDS combine LiDAR, multiple cameras, and SLAM over a walk-through to produce a Gaussian splat or mesh of a space. That output is a reconstruction. It is useful as a digital twin, but it is the result of a pipeline, not an observation of the world.

A world model trains on observations. It needs the raw pairing of appearance and geometry as the world presented it, frame by frame, so it can learn how the two relate. Liberty3D provides that pairing at 16K resolution, across the entire 360° view, with depth that was measured at capture rather than inferred later. The point clouds that come out of it are also a strong starting point for Gaussian splats and other reconstructions, for teams that want both.

How it is captured
Alia is a single camera that captures 360° in stereo in one shot. Depth is measured, not predicted, and it is aligned to the color image because both come from the same optics. The optical design is patented, with grants in multiple countries. The single-camera panoramic stereo method behind it was published at CVPR. DreamVu’s processing pipeline converts each capture into color, depth, and point cloud outputs without manual intervention.
“Everyone in this field is being offered data. What they are not being offered is real-world geometry captured at the same instant as the color, at a resolution where you can see the difference,” said Sashi Reddi, Co-Founder and CEO of DreamVu. “We built the camera, we built the pipeline, and Liberty3D is what comes out of it. Open any scene in 3D and judge it yourself. If it is what your model needs, we can capture wherever you need it.”

Availability
Selected Liberty3D scenes can be explored in the browser at dreamvu.ai/world-model-data, with color, depth, and 3D point cloud views of each. Access to the full corpus is available on request. DreamVu runs capture programs for customers in the environments their models need to learn from.

About DreamVu
DreamVu is an applied research company that captures the physical world in 3D. Founded on a patented single-camera 360° stereo design, the company delivers real-world capture data for training world models and generative video models, robot training data, and simulation environments. DreamVu is headquartered in Philadelphia, PA, with offices in Palo Alto, CA and research and development in Hyderabad, India.
Contact
Sanju Pillai
CMO, DreamVu
sanju@dreamvu.ai | +1 (267) 914-5213

Sanju Pillai
DreamVu
+1 267-914-5213
email us here
Visit us on social media:
LinkedIn

Media gallery