Summary
Mira-Scene reconstructs coherent, object-level 3D scenes from a single reference image. It jointly predicts object geometry and pixel-aligned canonical coordinates, aligning them with a scene-space point cloud to recover object placement.
Mira-Scene reconstructs coherent, object-level 3D scenes from a single reference image. It jointly predicts object geometry and pixel-aligned canonical coordinates, aligning them with a scene-space point cloud to recover object placement.
Developers can preserve object fidelity while assembling scenes for editing, visualization, and physical simulation. Core capabilities include Single-image scene reconstruction, Object-level geometry, Canonical Coordinate Maps, and Dense pixel correspondences.
Public code or model weights are linked from the project, with setup instructions for running or adapting the system. Additional characteristics include Scene-space point-cloud alignment, Compositional scene layouts, Blender simulation examples, and Isaac simulation demonstrations.
Get more likes & reach the top of search results by adding this button on your site!


Get top updates in AI to your inbox every weekend. It's free!