Students at the University of Toronto have published a pipeline that turns Meta Quest 3S scans into gaussian splats on a static website.
The project comes from the Mersivity Student Research Team, co-led by Alexander Vicol and Michel Herrera Viyella and working with Professor Steve Mann, the team told Radiance Fields. OpenQuestCapture began exporting passthrough imagery, poses and depth maps from the Quest 3 and 3S last December.
A Unity 6 app on the headset keeps a passthrough picture every 5 cm or 6 degrees of head movement, with the headset's position and direction and an 80 x 60 grid of depth rays. The Quest 3S has no lidar, so depth comes from its two tracking cameras. On a laptop, Brush trains the splat for 8,000 steps in about two minutes, using the headset poses directly with no structure from motion.
Open3D writes the trained splat as compressed .spz at about a tenth of the size, and Spark with three.js draws it in the browser. A second branch renders the splat's depth from every camera position, votes it into a grid to form a closed solid cut flat at its base, and slices it for a Bambu H2C without sending anything to a printer.
A live page turns each depth reading into one placed, untrained Gaussian per 2 cm of surface, arriving within a second through a relay that stores nothing. Only synthetic rooms have been sent to it, and the live view has not yet run from a headset. A walk around one object, AR and VR inside the headset, and a physical print remain undone.
The team's own Python, PowerShell and Unity C# tools are posted as plain files with no license stated. The pipeline is available now.





