| |
Lift4D is a new method for reconstructing complete 3D geometry, appearance, and deformation of dynamic objects from a single monocular video, including regions never seen by the camera. The approach combines a temporally consistent single-view 3D reconstruction model with deformable 3D Gaussian Splatting and a diffusion prior to handle challenging in-the-wild scenarios with severe occlusions and non-rigid motion. The method significantly outperforms prior 4D reconstruction approaches by leveraging both visual observations and data-driven priors throughout the reconstruction process.
Read Full Article →
← More Tech news