# Mind the Gap: Standard 3DGS Evaluation Primarily Measures Near-Trajectory Interpolation

arXiv: https://arxiv.org/abs/2607.01556 · HTML: https://mlsanity.com/paper/2607.01556/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 10 of its 57 references that are in the map.
- First posted on arXiv 2 Jul 2026.

## Builds on (most cited first, 10 of 10)

- [NerfAcc: A General NeRF Acceleration Toolbox](https://mlsanity.com/paper/2210.04847/) (2022)
- [InstantSplat: Sparse-view Gaussian Splatting in Seconds](https://mlsanity.com/paper/2403.20309/) (2024)
- [HY-World 2.0: A Multi-Modal World Model for Reconstructing, Generating, and Simulating 3D Worlds](https://mlsanity.com/paper/2604.14268/) (2026)
- [Novel View Extrapolation with Video Diffusion Priors](https://mlsanity.com/paper/2411.14208/) (2024)
- [PRIMU: Uncertainty Estimation for Novel Views in Gaussian Splatting from Primitive-Based Representations of Error and Coverage](https://mlsanity.com/paper/2508.02443/) (2025)
- [ExtraGS: Geometric-Aware Trajectory Extrapolation with Uncertainty-Guided Generative Priors](https://mlsanity.com/paper/2508.15529/) (2025)
- [Intrinsic Geometry-Appearance Consistency Optimization for Sparse-View Gaussian Splatting](https://mlsanity.com/paper/2603.02893/) (2026)
- [SLAM&Render: A Benchmark for the Intersection Between Neural Rendering, Gaussian Splatting and SLAM](https://mlsanity.com/paper/2504.13713/) (2025)
- [Do Flat Minima Improve Sparse Novel View Synthesis?](https://mlsanity.com/paper/2511.17918/) (2025)
- [Geo-EVS: Geometry-Conditioned Extrapolative View Synthesis for Autonomous Driving](https://mlsanity.com/paper/2604.07250/) (2026)

## Abstract (arXiv)

Standard MipNeRF360-style 3D Gaussian Splatting (3DGS) evaluation holds out every N-th frame -- but these frames have trained neighbors on both sides, so the metric measures near-trajectory interpolation rather than spatial generalization. We introduce a fair matched-count protocol that isolates this effect: both arms train on the same number of images and differ only in whether the holdout is spread evenly (interpolation) or forms a contiguous spatial sector (extrapolation). Our primary finding is a large, consistent interpolation-extrapolation gap of 3~12dB -- several times the differences typically reported between competing methods. The gap is robust to training noise, is in two cases large enough to flip a method ranking under multi-seed confirmation, and -- crucially -- persists across three representation families, including a non-Gaussian volumetric neural radiance field (NeRF), so it reflects spatial coverage rather than any one representation. Diagnostically, it is dominated by a diffuse/geometry-proxy component and tracks each view's angular distance to its nearest training view, a zero-cost signal that also guides capture planning; loss-side regularization yields only marginal gains. Standard holdouts remain useful for near-trajectory rendering but should not, alone, be read as evidence of spatial generalization. Prior work notes protocol sensitivity; ours is, to our knowledge, the first to combine matched-count paired holdout, cross-representation quantification, and a diagnostic analysis Table 1. We describe a spatial-holdout benchmark toolkit with standardized splits and baselines for 16 scenes, which we are preparing for public release.