02 / Featured

3D Gaussian Splatting Reconstruction Extension

Images and video → simulation-ready 3D you can rotate right here.

Input photos→Point cloud→3D Gaussian Splat→Mesh

[ p1_input_1.jpg ]waiting for asset

[ p1_input_2.jpg ]waiting for asset

[ p1_input_3.jpg ]waiting for asset

Step 1 — a handful of ordinary photos or a short video clip go in. Source: Tanks and Temples dataset (Knapitsch et al.)

[ p1_points.ply ]

Step 2 — structure-from-motion recovers a sparse 3D point cloud. Drag to orbit · scroll to zoom

[ p1_splat_viewer.html ]

Step 3 — a trained 3D Gaussian splat you can orbit in real time. Drag to orbit · scroll to zoom

[ p1_mesh.glb ]

Step 4 — a refined mesh, exported as USD/USDZ, ready for simulation. Drag to orbit · scroll to zoom

An Omniverse Kit extension that turns photos or video into 3D Gaussian splats and refined meshes — with text-prompt object isolation (SAM3), point-cloud-to-mesh tuning, and USD/USDZ export ready for simulation. Docker-orchestrated GPU backends (FVDB, COLMAP, Kaolin) with hardware validation and resumable pipelines. Iterated across six monthly releases.

Omniverse KitOpenUSDFVDBCOLMAPKaolinSAM3DockerPython

All projects