SM4RT: Learning Structured Motion Geometry for 4D Reconstruction
SM4RT models motion as shared rigid-body structure, not separate per-point flow.
The paper proposes a transformer that reconstructs 3D geometry and structured scene motion from monocular RGB video in one forward pass. Its “Structure-of-Motion” representation decomposes dynamics into compact motion bases, expressed as temporal SE(3) twists, with pixels assigned sparsely to those bases. The stated aim is to make points on the same object share a common motion trajectory, preserving kinematic structure while recovering dense motion. Code is listed as available. ArXiv · AI/CL/LG's note
The paper proposes a transformer that reconstructs 3D geometry and structured scene motion from monocular RGB video in one forward pass. Its “Structure-of-Motion” representation decomposes dynamics into compact motion bases, expressed as temporal SE(3) twists, with pixels assigned sparsely to those bases. The stated aim is to make points on the same object share a common motion trajectory, preserving kinematic structure while recovering dense motion. Code is listed as available. ArXiv · AI/CL/LG's note
score 5