GARDEN: Gravity-Aligned Reconstruction of Disentangled ENvironments from RGB images
Researchers have developed GARDEN, a new framework for reconstructing 3D environments from RGB images. This system uses gravity as a physical prior to disentangle rigid objects from background geometry, enabling direct physics simulation. Unlike previous methods that rely on CAD asset retrieval, GARDEN preserves scene-specific geometric fidelity and improves object placement reliability and rendering-simulation efficiency. AI
IMPACT Enables more realistic and interactive 3D environment generation for simulations.