cs.CVMar 12, 2026

Mobile-GS: Real-time Gaussian Splatting for Mobile Devices

Authors: Xiaobiao Du, Yida Wang, Kun Zhan, Xin Yu

Organizations: University of Technology Sydney · Li Auto Inc. · Adelaide University

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

3D Gaussian Splatting (3DGS) has emerged as a powerful representation for high-quality rendering across a wide range of applications.However, its high computational demands and large storage costs pose significant challenges for deployment on mobile devices. In this work, we propose a mobile-tailored real-time Gaussian Splatting method, dubbed Mobile-GS, enabling efficient inference of Gaussian Splatting on edge devices. Specifically, we first identify alpha blending as the primary computational bottleneck, since it relies on the time-consuming Gaussian depth sorting process. To solve this issue, we propose a depth-aware order-independent rendering scheme that eliminates the need for sorting, thereby substantially accelerating rendering. Although this order-independent rendering improves rendering speed, it may introduce transparency artifacts in regions with overlapping geometry due to the scarcity of rendering order. To address this problem, we propose a neural view-dependent enhancement strategy, enabling more accurate modeling of view-dependent effects conditioned on viewing direction, 3D Gaussian geometry, and appearance attributes. In this way, Mobile-GS can achieve both high-quality and real-time rendering. Furthermore, to facilitate deployment on memory-constrained mobile platforms, we also introduce first-order spherical harmonics distillation, a neural vector quantization technique, and a contribution-based pruning strategy to reduce the number of Gaussian primitives and compress the 3D Gaussian representation with the assistance of neural networks. Extensive experiments demonstrate that our proposed Mobile-GS achieves real-time rendering and compact model size while preserving high visual quality, making it well-suited for mobile applications.

Figures & tables

Figure 1: Mobile-GS is the first real-time Gaussian Splatting method that can reach 116 FPS rendering speed in the 1600 ×\times 1063 resolution for Bicycle on the mobile equipped with the Snapdragon 8 Gen 3 GPU as shown in (a) . We evaluate rendering quality, storage costs, and inference speed on an RTX 3090 Ti GPU in (b) and (c) . Our Mobile-GS integrates depth-aware order-independent rendering, compression, and distillation techniques to deliver comparable rendering quality compared with the original 3DGS, while substantially reducing the storage requirements to 4.8 MB and achieving 1098 FPS on the unbounded scene, thereby enabling efficient deployment on mobile devices.
Figure 2: Sorting as the primary performance bottleneck. Left: Runtime analysis of the original 3DGS highlights that the sorting operation incurs a significant computational overhead during inference. Right: Removing the sorting step substantially accelerates 3DGS, achieving several-fold speedup compared to the original implementation.
Figure 3: Rendering pipeline of our proposed Mobile-GS compared with 3DGS. In the inference stage, different from 3DGS, our proposed method eliminates the tile-based rendering and the 3D Gaussian sorting process typically required for accurate alpha blending. Instead, we first compute the color of each 3D Gaussian for its related pixels in parallel and accumulate the color value for each pixel. Then, we composite the foreground and background color in a single pass. To further improve performance and maintain visual quality, we propose a depth-aware order-independent rendering strategy that replaces the original sorting-dependent alpha blending.
Figure 4: Overall illustration and visualization of view-dependent opacity modeling. Left: We leverage an MLP fed with 3D Gaussian scale, rotation, spherical harmonics, and the vector of the camera toward the 3D Gaussian as input to predict a view-dependent opacity. Right: We display that our Mobile-GS removes redundant opacity and keeps important Gaussians with high opacity.
DatasetMip-NeRF360Tanks&TemplesDeep Blending
Method & MetricsPSNR ↑\uparrowSSIM ↑\uparrowLPIPS ↓\downarrowStorage ↓\downarrowFPS ↑\uparrowPSNR ↑\uparrowSSIM ↑\uparrowLPIPS ↓\downarrowStorage ↓\downarrowFPS ↑\uparrowPSNR ↑\uparrowSSIM ↑\uparrowLPIPS ↓\downarrowStorage ↓\downarrowFPS ↑\uparrow
3DGS ( Kerbl et al., 2023 )27.210.8150.214839.9 MB17423.140.8410.183371.5 MB23629.410.9030.243697.3 MB214
LightGaussian ( Fan et al., 2024 )27.080.8010.24460.4 MB22722.610.8030.26429.9 MB39228.740.8560.32548.2 MB271
AdR-Gaussian ( Wang et al., 2024a )26.950.7920.259358.2 MB25422.740.8090.251214.6 MB37228.920.8630.305251.4 MB284
SortFreeGS ( Hou et al., 2025 )27.020.7750.267851.4 MB73122.810.8170.254471.5 MB84828.690.8520.326724.2 MB793
Speedy-Splat ( Hanson et al., 2025 )26.920.7820.29679.4 MB40123.080.8210.24162.4 MB52729.110.8640.30971.2 MB463
Table 1: Quantitative comparisons of state-of-the-art 3D reconstruction methods on real-world datasets. We evaluate and report performance on three commonly used datasets, such as Mip-NeRF 360 ( Barron et al., 2022 ) , Tank&Temples ( Knapitsch et al., 2017 ) , and Deep Blending ( Hedman et al., 2018 ) . We highlight the best results among the lightweight 3DGS methods.
Figure 5: Qualitative comparisons of existing methods and our proposed Mobile-GS. We display the storage cost and FPS per scene to better demonstrate the performance of our method. We extract close-ups to highlight the differences.
Table 7
Figure 6: Runtime analysis of Mobile-GS.
Figure 7: Evaluation of the proposed neural view-dependent enhancement strategy. “w/o” and “w/” mean the removal and the integration of the proposed neural view-dependent enhancement strategy.
Table 10
Table 11
Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

MethodRenderingWeightingPSNR ↑\uparrowStorage ↓\downarrowFPS* ↑\uparrow
SortFreeGS*C=cbgwbg+∑i=1Nciαiw(di)wbg+∑i=1Nαiw(di)\mathbf{C}=\frac{c_{bg}w_{bg}+\sum_{i=1}^{\mathcal{N}}c_{i}\alpha_{i}w(d_{i})}{w_{bg}+\sum_{i=1}^{\mathcal{N}}\alpha_{i}w(d_{i})}w(di)=exp⁡ ⁣(−σdiβ)w(d_{i})=\exp\!\left(-\sigma d_{i}^{\beta}\right)26.7464.3 MB18
GESC=CsWs+CGWs+WG\mathbf{C}=\frac{C_{s}W_{s}+C_{G}}{W_{s}+W_{G}}WG(x^)=∑i=1N[\mathbbm1(di<ds(x^)+ϵ)]αi(x^)W_{G}(\hat{\mathbf{x}})=\sum_{i=1}^{N}[\mathbbm{1}(d_{i}<d_{s}(\hat{\mathbf{x}})+\epsilon)]\alpha_{i}(\hat{\mathbf{x}})27.0229.4 MB24
OursC=(1−T)∑i=1Nciαiwi∑i=1Nαiwi+Tcbg\mathbf{C}=\left(1-T\right)\frac{\sum_{i=1}^{\mathcal{N}}\mathbf{c}_{i}\alpha_{i}w_{i}}{\sum_{i=1}^{\mathcal{N}}\alpha_{i}w_{i}}+T\mathbf{c}_{bg}wi=ϕi2+ϕidi2+exp(smaxdi)w_{i}=\phi^{2}_{i}+\frac{\phi_{i}}{d_{i}^{2}}+exp(\frac{s_{max}}{d_{i}})27.124.6 MB127
Appendix
Table 8: Comparison with different sorting-free methods. SortFreeGS* means its quantized version. We report metrics on the Mip-NeRF 360 dataset for the mobile equipped with the Snapdragon 8 Gen 3 GPU. FPS* means the rendering speed on the mobile.
Figure 8: Additional visual comparisons of Speedy-Splat, SortFreeGS, 3DGS, and our proposed Mobile-GS. We highlight the close-up for better differentiation.
Figure 9: User study of rendering quality between our proposed Mobile-GS, 3DGS ( Kerbl et al., 2023 ) , and LocoGS-S ( Shin et al., 2025 ) . The higher rate means more users like it.
Method3DGSSpeedy-SplatSortFreeGSMobile-GS (Ours)
Cold-start FPS ↑\uparrow81924127
Steady‑state FPS ↑\uparrow3101874
Appendix
Table 12: Steady‑state FPS evaluation on the mobile. The larger FPS means faster rendering speed.
MethodPreprocessingSortingMLPRasterizationTotal
3DGS*1.642.0902.165.89
SortFreeGS*1.78002.254.03
Mobile-GS0.1700.240.420.83
Appendix
Table 13: Power draw measurement on the mobile. We measure on different Vulkan operators and report their power (W) on the mobile with the Snapdragon 8 Gen 3 GPU.
More in this topic

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

Explore similar work

CardsList
Jun 29, 2026

Monte Carlo Energy Aggregation for Mobile 3D Gaussian Splatting

Recent advances in 3D Gaussian Splatting have demonstrated unprecedented success in novel view synthesis. However, the substantial inference and storage overhead driven by high-order Spherical Harmonics (SH) are primary bottlenecks for mobile platforms. In this paper, we present Flux-GS, a real-time Gaussian Splatting method designed to achieve high-fidelity rendering with significantly reduced overhead for resource-constrained mobile platforms. We first propose a Monte Carlo Specular Energy Aggregator, sampling third-order radiance residuals and aggregating specular energy into a compact latent space. In this way, our method effectively preserves visually salient lighting features in lower-order bands without expensive distillation or pre-training. To mitigate the high-frequency details lost during compression, we introduce an Attribute-Conditioned SH Enhancement module. This module predicts Gaussian-aware offsets based on intrinsic Gaussian attributes, which enhance the first-order SH representation prior to inference, without extra inference costs. Furthermore, the original single-view gradient-based densification is prone to producing excessive Gaussians and overfitting to a certain view. We address these limitations by proposing a Multi-view Alpha-based Densification and Pruning strategy. By leveraging multi-view guidance, we ensure multi-view structure consistency and the precise removal of redundant primitives. Extensive experiments demonstrate that Flux-GS achieves substantial parameter reduction while maintaining competitive visual quality, offering a robust and scalable solution for real-time mobile rendering. Code: \textcolor{magenta}{https://xiaobiaodu.github.io/flux-gs-project/}.

Xiaobiao Du, YuAn Wang, Hao Li +3

University of Technology Sydney, Australia · Baidu Inc., China · Australian Institute for Machine Learning, Adelaide University, Australia

3DGS Compression3D Gaussian Splatting3DGS AccelerationAdaptive Gaussian Densification
AskPDF
Oct 4, 2026

Mobile-4DGS: Unified Static-Dynamic Real-time Mobile Gaussian Splatting

Recent advances in 3D Gaussian Splatting (3DGS) have achieved remarkable performance in novel view synthesis, yet deploying both static and dynamic Gaussian representations on resource-constrained mobile devices remains challenging due to heavy storage, redundant primitives, and costly per-frame computation. We present Mobile-4DGS, a unified lightweight framework for high-fidelity real-time static and dynamic Gaussian rendering on mobile platforms. For compact appearance modeling, we introduce a Monte Carlo Specular Energy Aggregator that compresses high-order radiance residuals into the first-order Spherical Harmonics (SH), together with an Attribute-Conditioned SH Enhancement module whose predicted offsets are pre-baked before inference. We further propose a Multi-View Alpha-Based Densification and Pruning strategy to suppress redundant primitives while maintaining multi-view consistency. For dynamic scenes, we develop a compact explicit 4D representation by constructing second-order Gaussian motion, learnable temporal support, and a binary static-dynamic partition, enabling continuous-time modeling without runtime deformation networks. Based on this partition, a Depth-Order Certificate selectively reuses previously committed depth orders to reduce re-projection, sorting, merging, and index-buffer updates during playback. Extensive experiments on static and dynamic scenes demonstrate that Mobile-4DGS substantially reduces storage and rendering overhead while maintaining competitive visual quality, enabling real-time 3D and 4D Gaussian Splatting on mobile devices. \textcolor{magenta}{Code has been released: https://xiaobiaodu.github.io/mobile-4dgs-project/}.

Xiaobiao Du, Zhen Fang, Tianqing Zhu +2

Adelaide University, Adelaide, 5005, Australia.

3DGS Compression3D Gaussian Splatting3DGS AccelerationDynamic Scene Reconstruction
AskPDF
Jan 24, 2026

PocketGS: High-Fidelity On-Device Training for 3D Gaussian Splatting

Although 3D Gaussian Splatting (3DGS) supports real-time rendering, training relies on offline geometry, long-horizon densification, and memory-intensive differentiation. We cast post-capture static-scene training as resource-constrained finite-horizon optimization, where terminal quality depends jointly on the initial representation and reference-consistent gradient execution. To the best of our knowledge, PocketGS is the first publicly documented pipeline to complete geometry construction, Gaussian initialization, and 3DGS optimization on a commodity smartphone. Its I, G warm start combines refined poses, a filtered dense prior, and surface-aligned anisotropic Gaussians. Its T operator replays compositing interactions and scatters gradients to canonical parameters, matching a numerical reference at the reported precision. Controlled 500-step experiments on LLFF, Mip-NeRF 360, and MobileScan show high-quality reconstruction at the device horizon. Across MobileScan, every scene remains below 3 GB peak memory and post-capture processing averages 255.2 s, enabling an on-device capture-to-render workflow. Separate 30k experiments characterize native-schedule behavior.

Wenzhi Guo, Guangchi Fang, Shu Yang +1

3D Gaussian Splatting3DGS Training3D Reconstruction3DGS Acceleration
AskPDF