4DAnyone Framework for 4D Human Reconstruction from Monocular Video
August 19, 2026
4DAnyone reconstructs 4D humans from uncalibrated monocular video by generating multiview-consistent videos and lifting them into 4D Gaussian Splatting. The method addresses bounded-attention-context bottlenecks in Diffusion Transformer (DiT) architectures.
HOW THIS AFFECTS YOU
●
researcherYou can improve 4DGS reconstruction quality by addressing view-consistency bottlenecks in DiT passes.
●
designerYou can generate high-fidelity 4D human models from simple single-camera video clips.