DOC-GS: Dual-Domain Observation and Calibration for Reliable Sparse-View Gaussian Splatting

Hantang Li; Qiang Zhu; Xiandong Meng; Debin Zhao; Xiaopeng Fan

cs.CVApr 8, 2026

DOC-GS: Dual-Domain Observation and Calibration for Reliable Sparse-View Gaussian Splatting

Authors: Hantang Li, Qiang Zhu, Xiandong Meng, Debin Zhao, Xiaopeng Fan

Organizations: Harbin Institute of Technology, Shenzhen, China · Pengcheng Laboratory, Shenzhen, China · Harbin Institute of Technology, Harbin, China

Abstract

Sparse-view reconstruction with 3D Gaussian Splatting (3DGS) is fundamentally ill-posed due to insufficient geometric supervision, often leading to severe overfitting and the emergence of structural distortions and translucent haze-like artifacts. While existing approaches attempt to alleviate this issue via dropout-based regularization, they are largely heuristic and lack a unified understanding of artifact formation. In this paper, we revisit sparse-view 3DGS reconstruction from a new perspective and identify the core challenge as the unobservability of Gaussian primitive reliability. Unreliable Gaussians are insufficiently constrained during optimization and accumulate as haze-like degradations in rendered images. Motivated by this observation, we propose a unified Dual-domain Observation and Calibration (DOC-GS) framework that models and corrects Gaussian reliability through the synergy of optimization-domain inductive bias and observation-domain evidence. Specifically, in the optimization domain, we characterize Gaussian reliability by the degree to which each primitive is constrained during training, and instantiate this signal via a Continuous Depth-Guided Dropout (CDGD) strategy, where the dropout probability serves as an explicit proxy for primitive reliability. This imposes a smooth depth-aware inductive bias to suppress weakly constrained Gaussians and improve optimization stability. In the observation domain, we establish a connection between floater artifacts and atmospheric scattering, and leverage the Dark Channel Prior (DCP) as a structural consistency cue to identify and accumulate anomalous regions. Based on cross-view aggregated evidence, we further design a reliability-driven geometric pruning strategy to remove low-confidence Gaussians.

Figures & tables

Figure 1. Motivation of DOC-GS for sparse-view Gaussian splatting. (a) To explore the degradation characteristics of rendered images, we generate the dark channel residual between the rendered image and ground-truth image. By comparison on the quantitative and qualitative results, the regions affected by floaters in the rendered images exhibit abnormal dark channel responses, which intuitively highlights artifact-prone regions. (b) Based on this insight, we reformulate sparse-view 3D Gaussian splatting as a dual-domain reliability inference problem. DOC-GS first introduces Continuous Depth-Guided Dropout (CDGD) to suppress unreliable Gaussians in the optimization domain, and then designs Dark Channel Prior-guided Geometric Pruning (DCP-GP) to detect and remove haze-like artifacts in the observation domain, ensuring high fidelity sparse-view reconstruction.

Figure 2. Overview of DOC-GS. Initial Gaussian primitives from sparse-view SfM are jointly optimized through two complementary domains. In the optimization domain, Continuous Depth-Guided Dropout (CDGD) introduces depth-aware uncertainty to suppress potentially unreliable Gaussians, improving optimization stability. In the observation domain, we leverage the Dark Channel Prior (DCP) to detect haze-like structural inconsistencies in rendered images. The resulting evidence is accumulated across views to estimate Gaussian reliability, which guides a reliability-driven geometric pruning strategy to remove persistently unreliable primitives. The entire process is performed iteratively, enabling progressive refinement of Gaussian representations and effective suppression of haze-like artifacts under sparse-view settings.

	Methods	3-view			6-view			9-view
	Methods	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
NeRF-based	Mip-NeRF ( Barron et al., 2021 )	16.11	0.401	0.460	22.91	0.756	0.213	24.88	0.826	0.170
	DietNeRF ( Jain et al., 2021 )	14.94	0.370	0.496	21.75	0.717	0.248	24.28	0.801	0.183
	RegNeRF ( Niemeyer et al., 2022 )	19.08	0.587	0.336	23.10	0.760	0.206	24.86	0.820	0.161
	FreeNeRF ( Yang et al., 2023 )	19.63	0.612	0.308	23.73	0.779	0.195	25.13	0.827	0.160
	SparseNeRF ( Wang et al., 2023 )	19.86	0.624	0.328	–	–	–	–	–	–

Table 1. Quantitative comparison on LLFF dataset under different sparse-view settings (3, 6, and 9 views). Colors indicate best , second best , and third best results.

Methods	12-view			24-view
Methods	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
3DGS ( Kerbl et al., 2023 )	18.52	0.523	0.415	22.80	0.708	0.276
FSGS ( Zhu et al., 2024 )	18.80	0.531	0.418	23.70	0.745	0.230
CoR-GS ( Zhang et al., 2024 )	19.52	0.558	0.418	23.39	0.727	0.271
DropGaussian ( Park et al., 2025 )	19.74	0.577	0.364	23.75	0.756	0.227
NexusGS ( Zheng et al., 2025 )	-	-	-	23.86	0.753	0.206

Table 2. Quantitative comparison on MipNeRF-360 dataset under different sparse-view setting (12 and 24 views).

Methods	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
3DGS ( Kerbl et al., 2023 )	21.56	0.847	0.130
DNGaussian ( Li et al., 2024 )	24.31	0.886	0.088
NexusGS ( Zheng et al., 2025 )	24.37	0.893	0.087
FSGS ( Zhu et al., 2024 )	24.64	0.895	0.095
CoR-GS ( Zhang et al., 2024 )	24.43	0.896	0.084
DropGaussian ( Park et al., 2025 )	25.42	0.888	0.089

Table 3. Quantitative comparison on Blender dataset under sparse-view setting (8 views).

Panel of Figure 3 — Figure 3. Qualitative comparison of our method and three existing methods on the LLFF ( Mildenhall et al., 2019 ) dataset under 3-view settings. Our method effectively suppresses floater artifacts and produces cleaner and more consistent geometric structures.

Panel of Figure 4 — Figure 4. Qualitative comparison of our method and three existing methods on the MipNeRF360 ( Barron et al., 2022 ) dataset under 12-view settings. Our method effectively suppresses floater artifacts and produces cleaner and more consistent geometric structures.

Methods	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
FSGS ( Zhu et al., 2024 )	20.43	0.682	0.248
FSGS + Ours	20.69	0.705	0.213
CoR-GS ( Zhang et al., 2024 )	20.45	0.712	0.196
CoR-GS + Ours	20.72	0.719	0.195
DropGaussian ( Park et al., 2025 )	20.76	0.713	0.200
DropGaussian + Ours	21.24	0.740	0.179

Table 4. Compatibility to existing sparse-view reconstruction methods with our design on the LLFF dataset (3-views).

Panel of Figure 5 — Figure 5. Compatibility to 3DGS variants on the LLFF dataset (3-views). The DOC-GS can be easily integrated into various 3DGS-based methods and improve their performance.

DDGS	CDGD	DCP-GP	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
$\times$	$\times$	$\times$	20.74	0.716	0.200
✓	$\times$	$\times$	20.94	0.722	0.196
$\times$	✓	$\times$	21.12	0.724	0.189
$\times$	$\times$	✓	21.19	0.733	0.184
$\times$	✓	✓	21.38	0.748	0.176

Table 5. Ablation study on the LLFF dataset (3 views).

3D Gaussian Splatting

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

See all 1,019 papers

May 12, 2026

PairDropGS: Paired Dropout-Induced Consistency Regularization for Sparse-View Gaussian Splatting

Dropout-based sparse-view 3D Gaussian Splatting (3DGS) methods alleviate overfitting by randomly suppressing Gaussian primitives during training. Existing methods mainly focus on designing increasingly sophisticated dropout strategies, while they overlook the resulting inconsistencies among different dropped Gaussian subsets. This oversight often leads to unstable reconstruction and suboptimal Gaussian representation learning.In this paper, we revisit dropout-based sparse-view 3DGS from a consistency regularization perspective and propose PairDropGS, a Paired Dropout-induced Consistency Regularization framework for sparse-view Gaussian splatting. Specifically, PairDropGS first constructs a pair of the dropped Gaussian subsets from a shared Gaussian field and designs a low-frequency consistency regularization to constrain their low-frequency rendered structures. This design encourages the shared Gaussian field to preserve stable scene layout and coarse geometry under different random dropouts, while avoiding excessive constraints on ambiguous high-frequency details. Moreover, we introduce a progressive consistency scheduling strategy to gradually strengthen the consistency regularization during training for stability and robustness of reconstruction. Extensive experiments on widely-used sparse-view benchmarks demonstrate that PairDropGS achieves superior training stability, significantly outperforms existing dropout-based 3DGS methods in reconstruction quality, while exhibiting the simplicity and plug-and-play nature for improving dropout-based optimization.

Hantang Li, Qiang Zhu, Xiandong Meng +3

School of Computer Science, Harbin Institute of Technology, Shenzhen, China · Pengcheng Laboratory, Shenzhen, China · Smart Coding Institute, Pengcheng Laboratory, Shenzhen, China +1

3D Gaussian Splatting Sparse-View 3D Reconstruction Consistency Regularization

Ask PDF

Sep 19, 2026

D3GS: Depth, DINO, and RGB Diffusion Co-Guided 3D Gaussian Splatting for Sparse-View Reconstruction

Novel view synthesis from sparse inputs remains challenging for 3D Gaussian Splatting (3DGS) due to ambiguous geometry, cross-view inconsistency, and missing details in under-constrained regions, resulting in degraded reconstruction and unstable rendering. To tackle these issues, we propose D

^{3}

GS, a Depth-DINO-Diffusion guided sparse-view Gaussian reconstruction framework that jointly enhances geometry and appearance. D

^{3}

GS first recovers a high-resolution, metric depth map via diffusion-based completion and DPT (Dense Prediction Transformer) refinement, providing robust Gaussian initialization and geometric constraints. Then, a DINO-guided view-consistent learning is introduced to augment Gaussian attributes with structural features, improving multi-view consistency. Finally, a diffusion-based Gaussian refinement module injects generative priors into an iterative optimization strategy, enhancing high-frequency geometric and appearance details within the Gaussian representation. Experiments on DTU, LLFF, and Mip-NeRF 360 show that D

^{3}

GS achieves consistent and substantial improvements over strong baselines, with ablation studies validating the effectiveness and complementary roles of each component.

Yunqi Gao, Zhanfeng Liao, Hanzhang Tu +7

Central China Normal University · Tsinghua University · Fuzhou University +4

Sparse-View 3D Reconstruction 3D Gaussian Splatting 3D Reconstruction Multi-View Consistency

Ask PDF

Mar 1, 2026

HeroGS: Hierarchical Guidance for Robust 3D Gaussian Splatting under Sparse Views

3D Gaussian Splatting (3DGS) has recently emerged as a promising approach in novel view synthesis, combining photorealistic rendering with real-time efficiency. However, its success heavily relies on dense camera coverage; under sparse-view conditions, insufficient supervision leads to irregular Gaussian distributions, characterized by globally sparse coverage, blurred background, and distorted high-frequency areas. To address this, we propose HeroGS, Hierarchical Guidance for Robust 3D Gaussian Splatting, a unified framework that establishes hierarchical guidance across the image, feature, and parameter levels. At the image level, sparse supervision is converted into pseudo-dense guidance, globally regularizing the Gaussian distributions and forming a consistent foundation for subsequent optimization. Building upon this, Feature-Adaptive Densification and Pruning (FADP) at the feature level leverages low-level features to refine high-frequency details and adaptively densifies Gaussians in background regions. The optimized distributions then support Co-Pruned Geometry Consistency (CPG) at parameter level, which guides geometric consistency through parameter freezing and co-pruning, effectively removing inconsistent splats. The hierarchical guidance strategy effectively constrains and optimizes the overall Gaussian distributions, thereby enhancing both structural fidelity and rendering quality. Extensive experiments demonstrate that HeroGS achieves high-fidelity reconstructions and consistently surpasses state-of-the-art baselines under sparse-view conditions.

Jiashu Li, Xumeng Han, Zhaoyang Wei +5

University of the Chinese Academy of Sciences

View-Dependent 3DGS 3D Gaussian Splatting Sparse-View 3D Reconstruction Adaptive Gaussian Densification 3D Gaussian Pruning

Ask PDF

Abstract

3D Gaussian Splatting

Explore similar work

PairDropGS: Paired Dropout-Induced Consistency Regularization for Sparse-View Gaussian Splatting

D3GS: Depth, DINO, and RGB Diffusion Co-Guided 3D Gaussian Splatting for Sparse-View Reconstruction

HeroGS: Hierarchical Guidance for Robust 3D Gaussian Splatting under Sparse Views