How do you handle occlusions and temporal consistency when reconstructing a dynamic 3D scene from a monocular camera stream?
