RoGe: Novel View Synthesis via End-to-End Implicit Reconstruction and Generation
RoGe creates a unified framework for novel view synthesis by linking an implicit scene reconstruction model with a video generation model end-to-end. It builds an implicit scene representation from sparse views, extracts per-view geometric features by querying this representation with camera rays, and conditions a diffusion model directly on these features. This joint training allows the generation process to refine its own geometric understanding for superior quality.