VolSplat: Rethinking Feed-Forward 3D Gaussian Splatting with Voxel-Aligned Prediction

Weijie Wang1,2*  Yeqing Chen3*  Zeyu Zhang2  Hengyu Liu2,4  Haoxiao Wang1  Zhiyuan Feng5 
Wenkang Qin2  Zheng Zhu2†  Donny Y. Chen6  Bohan Zhuang1†

1Zhejiang University   2GigaAI   3University of Electronic Science and Technology of China   
4The Chinese University of Hong Kong   5Tsinghua University   6Monash University

* Equal contribution † Corresponding authors

Paper

Code

Models

TL;DR

VolSplat improves multi-view consistency and geometric accuracy for feed-forward 3DGS with voxel-aligned prediction.



Pixel-aligned feed-forward 3DGS methods suffer from two primary limitations: 1) 2D feature matching struggles to effectively resolve the multi-view alignment problem, and 2) the Gaussian density is constrained and cannot be adaptively controlled according to scene complexity. We propose VolSplat, a framework that directly regresses Gaussians from 3D features based on a voxel-aligned prediction strategy. This approach achieves adaptive control over scene complexity and resolves the multi-view alignment challenge.

Architecture



Overview of our VolSplat. Given multi-view images as input, we first extract 2D features for each image using a Transformer-based network and construct per-view cost volumes with plane sweeping. Depth Prediction Module then estimates a depth map for each view, which is used to unproject the 2D features into 3D space to form a voxel feature grid. Subsequently, we employ a sparse 3D decoder to refine these features in 3D space and predict the parameters of a 3D Gaussian for each occupied voxel. Finally, novel views are rendered from the predicted 3D Gaussians.

Comparisons with SoTA Models

Visualization on RealEstate10K
GNN MVSplat TransSplat DepthSplat VolSplat (Ours) Ground Truth

Comparisons of Geometry Quality

Gaussians of DepthSplat and VolSplat
DepthSplat VolSplat (Ours)

Comparisons of Cross-dataset Generalization

Visualization on RealEstate10K → ACID
GNN MVSplat TransSplat DepthSplat VolSplat (Ours) Ground Truth

Citation

@article{wang2025volsplat,
  title={VolSplat: Rethinking Feed-Forward 3D Gaussian Splatting with Voxel-Aligned Prediction},
  author={Wang, Weijie and Chen, Yeqing and Zhang, Zeyu and Liu, Hengyu and Wang, Haoxiao and Feng, Zhiyuan and Qin, Wenkang and Zhu, Zheng and Chen, Donny Y. and Zhuang, Bohan},
  journal={arXiv preprint arXiv:2509.19297},
  year={2025}
}