Holo4D, described in a new arXiv preprint, addresses the problem of synthesizing a monocular video from a new camera trajectory. The authors note that this requires preserving the observed scene geometry and motion while completing regions that become newly visible.
The proposed approach is a holistic 4D reconstruction method that uses geometric control for video diffusion. Feed-forward 4D reconstructors recover the structure of the scene, and the geometric control guides the diffusion process to fill in unseen areas.
The abstract is truncated, so details on evaluation and comparisons are not available in the source. The paper's contribution appears to be tying 4D reconstruction more directly to the generative video-diffusion pipeline.