🚀 Lift4D: 4D Reconstruction from a Single Video
A framework called Lift4D has been developed that allows for the creation of dynamic 4D object models using only a single monocular video. The system combines deformable Gaussians (Gaussian Splatting) with diffusion priors, which helps reconstruct geometry even in occluded areas.
🌍 This solves the problem of data scarcity for training 4D models, allowing for the creation of high-quality 3D content without the use of specialized depth sensors.
👤 It is now possible to create full volumetric models of objects simply by filming them with a standard smartphone camera, even if they are constantly moving.
Source 1: https://lift4d.github.io/ Source 2: https://arxiv.org/abs/2606.23688