OctreeSplat: Hierarchical Octree-Guided Feed-Forward Gaussian Splatting

Yibin Zhao; Yihan Pan; Wenli Yang; Jun Nan; Jianjun Yi

cs.CVNov 2, 2026

OctreeSplat: Hierarchical Octree-Guided Feed-Forward Gaussian Splatting

Authors: Yibin Zhao, Yihan Pan, Wenli Yang, Jun Nan, Jianjun Yi

Organizations: East China University of Science and Technology, Shanghai, China · Shanghai Open University, Shanghai, China

Abstract

3D Gaussian Splatting (3DGS) has shown considerable potential for novel view synthesis (NVS) and 3D reconstruction. Feed-forward methods avoid costly per-scene optimization, but fixed image or spatial resolutions limit their ability to allocate primitives according to local complexity. We present OctreeSplat, a hierarchical octree-guided feed-forward Gaussian splatting. OctreeSplat extracts multi-view features and geometric priors from a frozen visual foundation model (VFM), constructs a hierarchical octree, and predicts densification scores. The resulting representation allocates Gaussian primitives at spatially varying densities according to local complexity. Stochastic solids volume rendering improves depth rendering, strengthens geometric supervision, and supports surface reconstruction. When camera parameters are available, an optional test-time optimization (TTO) module improves agreement between predicted and ground-truth parameters. Training updates only 76 million additional parameters and requires 35 GPU-hours. Experiments on Replica, TUM, Tanks and Temples, and DTU demonstrate competitive RGB view synthesis, depth synthesis, and surface reconstruction. OctreeSplat reduces the number of Gaussian primitives while maintaining high reconstruction quality.

Figures & tables

Fig. 2: Overview of the OctreeSplat pipeline. A frozen visual foundation model extracts multi-view features and predicts geometry. DPT and SPUNet construct multi-resolution voxel features and predict Gaussian parameters and densification scores, which guide adaptive octree subdivision to construct an octree-aligned Gaussian representation that adapts to local scene complexity. Stochastic-solid volume rendering provides RGB and geometric supervision.

Fig. 3: Improvement in pose estimation accuracy with TTO.

Fig. 4: Comparison of the maximum multi-view sampling rate and predicted densification scores.

Fig. 5: Depth maps rendered using standard 3DGS and stochastic-solid volume rendering for the same Gaussian scene.

Panel of Figure 5 — Fig. 6: Qualitative comparison of RGB view synthesis.

Panel of Figure 6 — Fig. 7: Qualitative comparison of depth map synthesis.

DTU
Method	Trans. (mm) $\downarrow$		Rot. ( ∘ ) $\downarrow$
Method	Mean	Median	Mean	Median
YoNoSplat	16.383	11.687	5.562	5.210
AnySplat	5.846	4.581	2.105	2.036
QuerySplat	3.043	2.773	0.971	0.969
OctreeSplat & DA3	2.007	1.535	0.530	0.511

TABLE III: Quantitative pose estimation results with 20 input views.

Replica
Method	Before ICP			After ICP
Method	Acc.	Com.	CD	Acc.	Com.	CD
AnySplat	51.905	49.643	50.774	41.793	38.600	40.196
DA3	65.833	64.426	65.129	46.325	43.459	44.892
YoNoSplat	48.781	46.121	47.451	43.176	40.751	41.963
QuerySplat	36.569	34.523	35.546	33.769	31.730	32.750

TABLE IV: Surface reconstruction results before and after ICP alignment. All distances are measured in mm.

Fig. 8: Qualitative comparison of reconstructed surfaces.

TABLE V: Ablation studies. (a) Stochastic-solid volume rendering. (b) Depth supervision. (c) Voxel size. (d) Octree subdivision.

Fig. 10: Spatial density of Gaussian primitives. Redder regions indicate higher primitive density.

3D Gaussian Splatting

3D Gaussian Splatting represents scenes as optimizable 3D Gaussian primitives for fast, photorealistic rendering from new viewpoints.

See all 1,043 papers

ViewSplat: View-Adaptive 3D Gaussian Splatting for Feed-Forward Synthesis
Mar 26, 2026Moonyeon Jeong, Seunggi Min, Suhyeon Lee +13D Gaussian Splatting Novel View Synthesis
University of Seoul · Korea Electronics Technology Institute
ATSplat: Compact Feed-forward 3D Gaussian Splatting with Adaptive Token Expansion
Jul 22, 2026In Cho, Jeonghwan Cho, Mijin Yoo +23D Gaussian Splatting Adaptive Gaussian Densification
Yonsei University, South Korea · National University of Singapore, Singapore
GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens
Apr 16, 2026Roni Itkin, Noam Issachar, Yehonatan Keypur +33D Gaussian Splatting 3DGS Acceleration
The Hebrew University of Jerusalem · Westlake University

Mar 26, 2026J/K move · Enter open · S save

ViewSplat: View-Adaptive 3D Gaussian Splatting for Feed-Forward Synthesis

Moonyeon Jeong, Seunggi Min, Suhyeon Lee, Hongje Seong

University of Seoul · Korea Electronics Technology Institute

3D Gaussian Splatting Novel View Synthesis Feed-Forward 3D Reconstruction View-Dependent 3DGS

Open paper AskPDF arXiv

We present ViewSplat, a view-adaptive 3D Gaussian splatting network for novel view synthesis from unposed images. While recent feed-forward 3D Gaussian splatting has significantly accelerated 3D scene reconstruction by bypassing per-scene optimization, a fundamental fidelity gap remains. We attribute this gap to the limited capacity of single-step feed-forward networks to regress static Gaussian primitives that satisfy all viewpoints. To address this limitation, we shift the paradigm from static primitive regression to view-adaptive splatting. Instead of a rigid Gaussian representation, our pipeline learns a view-adaptive latent representation. Specifically, ViewSplat initially predicts base Gaussian primitives alongside the weights of scene-conditioned View MLPs. During rendering, these MLPs take target-view coordinates as input and predict view-dependent residual updates for each Gaussian attribute (i.e., 3D position, scale, rotation, opacity, and color). This mechanism, which we term view-adaptive splatting, allows each primitive to rectify initial estimation errors, effectively capturing high-fidelity appearances. Extensive experiments demonstrate that ViewSplat achieves state-of-the-art fidelity while maintaining fast inference and real-time rendering; our large backbone variant runs at 15 FPS during inference and 90 FPS during rendering. Our project page is available at https://cvlab-uos.github.io/ViewSplat.

Abstract

3D Gaussian Splatting

Explore similar work

ViewSplat: View-Adaptive 3D Gaussian Splatting for Feed-Forward Synthesis

ATSplat: Compact Feed-forward 3D Gaussian Splatting with Adaptive Token Expansion

GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens