🔮OracleGS: Grounding Generative Priors
for Sparse-View Gaussian Splatting

Atakan Topaloğlu1,2,3, Kunyi Li4,6, Michael Niemeyer5,
Nassir Navab4,6, A. Murat Tekalp2,3, Federico Tombari4,5,6,

1 ETH Zürich, 2 Koç University, 3 KUIS AI Center, 4 Technical University of Munich,
5 Google, 6 Munich Center for Machine Learning
WACV 2026 (Oral)
Paper Supplementary arXiv Code Video
OracleGS Teaser

OracleGS reconciles generative completeness with regressive fidelity for sparse-view Gaussian Splatting, achieving state-of-the-art novel view synthesis.

Abstract

Sparse-view novel view synthesis is fundamentally ill-posed due to severe geometric ambiguity. Current methods are caught in a trade-off: regressive models are geometrically faithful but incomplete, whereas generative models can complete scenes but often introduce structural inconsistencies. We propose OracleGS, a novel framework that reconciles generative completeness with regressive fidelity for sparse view Gaussian Splatting. Instead of using generative models to patch incomplete reconstructions, our "propose-and-validate" framework first leverages a pre-trained 3D-aware diffusion model to synthesize novel views to propose a complete scene. We then repurpose a multi-view stereo (MVS) model as a 3D-aware oracle to validate the 3D uncertainties of generated views, using its attention maps to reveal regions where the generated views are well-supported by multi-view evidence versus where they fall into regions of high uncertainty due to occlusion, lack of texture, or direct inconsistency. This uncertainty signal directly guides the optimization of a 3D Gaussian Splatting model via an uncertainty-weighted loss. Our approach conditions the powerful generative prior on multi-view geometric evidence, filtering hallucinatory artifacts while preserving plausible completions in under-constrained regions, outperforming state-of-the-art methods on datasets including Mip-NeRF 360 and NeRF Synthetic.

Method Overview

SVG mit img laden

Our method consists of three main steps:

  1. Given sparse input views with poses, we first estimate initial point cloud and depth maps.
  2. Afterwards, a 3D-Aware generative model proposes novel synthetic views, while the 3D-Aware Oracle’s attention maps are used as a proxy for 3D uncertainty.
  3. Finally, we train the 3DGS model using a standard loss on the GT views and our novel uncertainty-guided loss on the synthetic views. We employ a progressive augmentation strategy over the course of the optimization to control the ratio of GT and synthetic images at each iteration, which helps to stabilize training and guide scene structure.

Visual Comparison with the State-of-the-Art

Comparison on Mip-NeRF360

Our method achieves state-of-the-art performance on the Mip-NeRF360 benchmark.

CoR-GS [ECCV'24] Ours DropGaussian [CVPR'25] Ours CoR-GS [ECCV'24] Ours

Comparison on NeRF Synthetic

We achieve the highest SSIM and LPIPS among all the baselines on the NeRF Synthetic benchmark.

CoR-GS [ECCV'24] Ours DropGaussian [CVPR'25] Ours

Interactive Demo

Explore one of our reconstructed scenes in real-time.

OracleGS on Kitchen scene from the Mip-NeRF360 dataset in the 24-view setting.

BibTeX

@InProceedings{Topaloglu_2026_WACV,
      author    = {Topalo\u{g}lu, Atakan and Li, Kunyi and Niemeyer, Michael and Navab, Nassir and Tekalp, A. Murat and Tombari, Federico},
      title     = {OracleGS: Grounding Generative Priors for Sparse-View Gaussian Splatting},
      booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
      month     = {March},
      year      = {2026},
      pages     = {77-87}
  }

Acknowledgements

The results we show above are from the Mip-NeRF360 and the NeRF Synthetic datasets. The website is built on top of the Nerfies template and uses the image slider.