# Frame-Consistent Feed-Forward 3D Gaussian Splatting from Sparse Unposed Images

arXiv: https://arxiv.org/abs/2610.23039 · HTML: https://mlsanity.com/paper/2610.23039/

## Key facts

- No citing papers in the map yet (data through 19 Oct 2026).
- Placed in 3D Gaussian splatting methods, one of 15 directions in 3D reconstruction and neural rendering.
- Builds on 22 of its 29 references that are in the map.
- First posted on arXiv 16 Oct 2026.

## Placed in

- 3D reconstruction and neural rendering / [3D Gaussian splatting methods](https://mlsanity.com/recon3d/3d-gaussian-splatting-methods/)

## Builds on (most cited first, 22 of 22)

- [The Unreasonable Effectiveness of Deep Features as a Perceptual Metric](https://mlsanity.com/paper/1801.03924/) (2018)
- [3D Gaussian Splatting for Real-Time Radiance Field Rendering](https://mlsanity.com/paper/2308.04079/) (2023)
- [Mip-NeRF 360: Unbounded Anti-Aliased Neural Radiance Fields](https://mlsanity.com/paper/2111.12077/) (2021)
- [DUSt3R: Geometric 3D Vision Made Easy](https://mlsanity.com/paper/2312.14132/) (2023)
- [VGGT: Visual Geometry Grounded Transformer](https://mlsanity.com/paper/2503.11651/) (2025)
- [Grounding Image Matching in 3D with MASt3R](https://mlsanity.com/paper/2406.09756/) (2024)
- [Scaffold-GS: Structured 3D Gaussians for View-Adaptive Rendering](https://mlsanity.com/paper/2312.00109/) (2023)
- [DL3DV-10K: A Large-Scale Scene Dataset for Deep Learning-based 3D Vision](https://mlsanity.com/paper/2312.16256/) (2023)
- [LightGaussian: Unbounded 3D Gaussian Compression with 15x Reduction and 200+ FPS](https://mlsanity.com/paper/2311.17245/) (2023)
- [DepthSplat: Connecting Gaussian Splatting and Depth](https://mlsanity.com/paper/2410.13862/) (2024)
- [No Pose, No Problem: Surprisingly Simple 3D Gaussian Splats from Sparse Unposed Images](https://mlsanity.com/paper/2410.24207/) (2024)
- [AnySplat: Feed-forward 3D Gaussian Splatting from Unconstrained Views](https://mlsanity.com/paper/2505.23716/) (2025)
- [Mini-Splatting: Representing Scenes with a Constrained Number of Gaussians](https://mlsanity.com/paper/2403.14166/) (2024)
- [Visual Geometry Grounded Deep Structure From Motion](https://mlsanity.com/paper/2312.04563/) (2023)
- [MASt3R-SfM: a Fully-Integrated Solution for Unconstrained Structure-from-Motion](https://mlsanity.com/paper/2409.19152/) (2024)
- [PF3plat: Pose-Free Feed-Forward 3D Gaussian Splatting](https://mlsanity.com/paper/2410.22128/) (2024)
- [SelfSplat: Pose-Free and 3D Prior-Free Generalizable 3D Gaussian Splatting](https://mlsanity.com/paper/2411.17190/) (2024)
- [TokenGS: Decoupling 3D Gaussian Prediction from Pixels with Learnable Tokens](https://mlsanity.com/paper/2604.15239/) (2026)
- [TokenSplat: Token-aligned 3D Gaussian Splatting for Feed-forward Pose-free Reconstruction](https://mlsanity.com/paper/2603.00697/) (2026)
- [3R-GS: Best Practice in Optimizing Camera Poses Along with 3DGS](https://mlsanity.com/paper/2504.04294/) (2025)
- [VGGT-X: When VGGT Meets Dense Novel View Synthesis](https://mlsanity.com/paper/2509.25191/) (2025)
- [An Evaluation of DUSt3R/MASt3R/VGGT 3D Reconstruction on Photogrammetric Aerial Blocks](https://mlsanity.com/paper/2507.14798/) (2025)

## Abstract (arXiv)

Feed-forward Gaussian models have made novel view synthesis almost instantaneous. Feed-forward methods predict Gaussians directly, either in two stages or in a single merged pass. Using two separately trained models is attractive because each can specialize in its own objective: one for accurate pose estimation and the other for high-quality reconstruction, rather than forcing a single jointly trained network to compromise between the two. A natural instantiation pairs a frozen pose estimator, VGGT, with a frozen reconstruction backbone, TokenGS. Prior work, such as VGGT-X and 3R-GS, also adopts a two-stage design, but attributes the resulting quality degradation to noisy pose estimates and introduces learned refinement. We show that this is not the primary source of the problem. Through per-camera analysis across 107 scenes from two datasets, DL3DV and MipNeRF360, using COLMAP poses as the reference, we find that the rotation error after alignment is small (median 1.6 degrees), whereas the translation scale is offset by a large, dataset-consistent factor. Controlled ablations over 140 DL3DV scenes show that the primary obstacles are two integration failures rather than pose error: a coordinate-frame inconsistency and a translation-scale mismatch, both of which can be corrected deterministically. With these fixes, a frozen VGGT driving a frozen TokenGS achieves 22.47 PSNR, within 0.51 dB of the COLMAP reference, with no additional training, when poses are estimated from all scene views. In the sparse setting, where reconstruction uses only four context images under the standard pose-free protocol, the same pipeline matches this all-view quality, yielding a fully COLMAP-free pipeline without any refinement step.