We propose scene-adaptive strategies to efficiently allocate representation capacity for generating immersive experiences of indoor environments from incomplete observations. Indoor scenes with multiple rooms often exhibit irregular layouts with varying complexity, containing clutter, occlusion, and flat walls. We maximize the utilization of limited resources with guidance from geometric priors, which are often readily available after pre-processing stages. We record observation statistics on the estimated geometric scaffold and guide the optimal placement of bases, which greatly improves upon the uniform basis arrangements adopted by previous scalable scene representations. We also suggest scene-adaptive virtual viewpoints to compensate for geometric deficiencies inherent in view configurations in the input trajectory and impose the necessary regularization. We present a comprehensive analysis and discussion regarding rendering quality and memory requirements in several large-scale indoor scenes, demonstrating significant enhancements compared to baselines that employ regular placements.
Method Overview. (a) We first extract the geometric scaffold from multi-view image observation. (b) Then we define coverage weights $w_i$ for surface points $\mathbf{x}_i$ on the obtained mesh surface, which consider both scene geometry and observation statistics. Starting from bases sampled along the camera trajectory using the FPS algorithm, we optimize their positions by minimizing the energy function $\mathcal{L}_{\text{cov}}$. (c) Finally, we optimize radiance fields with RGB supervision, guided by geometric regularization on training and virtual viewpoints.
Free-viewpoint rendering on the Zip-NeRF NYC scene. By applying our geometry-aware scene configuration to NeLF-Pro and LocalRF, we demonstrate consistent improvements in rendering quality. Results generated from just 250 views.
| GT | Nerfacto | Depth-Nerfacto | 3DGS | NeLF-Pro | Ours |
|---|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Qualitative results of novel view synthesis on ScanNet++ dataset. We compare our framework against NeRFacto, depth-regularized NeRFacto, 3DGS, and NeLF-Pro under severe extrapolation scenario.
| GT | Mega-NeRF | NeLF-Pro | NeLF-Pro + Ours | LocalRF | LocalRF + Ours |
|---|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Qualitative results of novel view synthesis on Zip-NeRF dataset. We compare our method against Mega-NeRF, NeLF-Pro, and LocalRF which leverage multiple bases for large-scale indoor scenes.
Qualitative comparison on ScanNet++. Our method outperforms state-of-the-art 3D Gaussian Splatting baselines under the sparse-view setting, producing sharper textures and fewer artifacts in challenging indoor scenes.
Qualitative comparison on Zip-NeRF. Our geometry-aware scene configuration consistently improves rendering quality over 3DGS baselines on large-scale unbounded outdoor scenes, demonstrating generalization beyond indoor environments.
@article{kim2025geometry,
title={Geometry-Aware Scene Configurations for Novel View Synthesis},
author={Kim, Minkwan and Choi, Changwoon and Kim, Young Min},
journal={arXiv preprint arXiv:2510.09880},
year={2025}
}