MetaView: Monocular Novel View Synthesis with Scale-Aware Implicit Geometry Priors

MetaView is a diffusion-based framework for monocular novel view synthesis from a single image, targeting large viewpoint changes. It combines implicit geometry priors from a feed-forward geometry perception network with metric depth, aiming to preserve the flexibility of generative scene modeling while improving geometric consistency and camera controllability. The authors report that MetaView outperforms existing methods on challenging monocular large-viewpoint-change benchmarks and generalizes better. Implementation code is publicly available through KlingAIResearch’s GitHub repository.