ImprovedGS+: A High-Performance C++/CUDA Re-Implementation Strategy for 3D Gaussian Splatting

Jordi Muñoz Vicente

cs.CVMar 9, 2026

ImprovedGS+: A High-Performance C++/CUDA Re-Implementation Strategy for 3D Gaussian Splatting

Authors: Jordi Muñoz Vicente

Organizations: Universidad de Murcia Murcia, Spain

Abstract

Recent advancements in 3D Gaussian Splatting (3DGS) have shifted the focus toward balancing reconstruction fidelity with computational efficiency. In this work, we propose ImprovedGS+, a high-performance, low-level reinvention of the ImprovedGS strategy, implemented natively within the LichtFeld-Studio framework. By transitioning from high-level Python logic to hardware-optimized C++/CUDA kernels, we achieve a significant reduction in host-device synchronization and training latency. Our implementation introduces a Long-Axis-Split (LAS) CUDA kernel, custom Laplacian-based importance kernels with Non-Maximum Suppression (NMS) for edge scores, and an adaptive Exponential Scale Scheduler. Experimental results on the Mip-NeRF360 dataset demonstrate that ImprovedGS+ establishes a new Pareto-optimal front for scene reconstruction. Our 1M-budget variant outperforms the state-of-the-art MCMC baseline by achieving a 26.8% reduction in training time (saving 17 minutes per session) and utilizing 13.3% fewer Gaussians while maintaining superior visual quality. Furthermore, our full variant demonstrates a 1.28 dB PSNR increase over the ADC baseline with a 38.4% reduction in parametric complexity. These results validate ImprovedGS+ as a scalable, high-speed solution that upholds the core pillars of Speed, Quality, and Usability within the LichtFeld-Studio ecosystem.

Figures & tables

Panel of Figure 1 — Figure 1: Visual comparison of edge importance maps on the Bicycle scene. The zoomed regions highlight how our CUDA-based NMS kernel reduces the noise and focuses in a thinned structural backbone compared to the baseline.

Scene	Garden			Bonsai
	PSNR $\uparrow$	SSIM $\uparrow$	$\#G(10^{6})$ $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	$\#G(10^{6})$ $\downarrow$
IGS (Baseline)	26.86	0.845	1.310	30.39	0.935	0.535
IGS+ (Ours)	26.08	0.809	1.400	30.34	0.932	0.573
-w/o scale scheduler	25.95	0.806	1.400	30.18	0.933	0.575

Table 1: Bonsai and Garden scenes Ablations table at 7,000 iterations for ImprovedGS+. Averaged from 6 independent runs.

Scene	Garden			Bonsai
	PSNR $\uparrow$	SSIM $\uparrow$	$\#G(10^{6})$ $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	$\#G(10^{6})$ $\downarrow$
IGS (Baseline)	29.05	0.887	2.614	33.16	0.953	0.827
IGS+ (Ours)	29.39	0.896	2.930	33.37	0.954	0.830
-w/o scale scheduler	29.16	0.894	2.937	32.95	0.952	0.843

Table 2: Bonsai and Garden scenes Ablations table at 30,000 iterations for ImprovedGS+. Averaged from 6 independent runs.

Dataset	MipNerf360
	PSNR $\uparrow$	SSIM $\uparrow$	Time $\downarrow$	$\#G(10^{6})$ $\downarrow$
MCMC 1M	28.64	0.848	62m14s	1.000
Ours (IGS+ 1M)	28.79	0.859	45m33s	0.867
Ours (IGS+)	29.38	0.877	62m10s	1.688
ADC	28.10	0.848	76m02s	2.741

Table 3: Experimental Results across the Mip-Nerf360 dataset at 30,000 iterations. Metrics for each scene are averaged from 6 independent runs. For a fair comparative analysis, the best results between MCMC 1M and ImprovedGS+ 1M are highlighted in red.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Scene	Bonsai				Counter
	PSNR $\uparrow$	SSIM $\uparrow$	Time $\downarrow$	#G $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	Time $\downarrow$	#G $\downarrow$
MCMC 1M	32.91	0.951	5m54s	1.000	30.84	0.928	6m48s	1.000
ImprovedGS+	33.37	0.954	4m35s	0.830	31.05	0.929	4m13s	0.618
ADC	32.10	0.942	4m12s	1.147	29.19	0.897	4m08s	0.875
Scene	Kitchen				Room
	PSNR $\uparrow$	SSIM $\uparrow$	Time $\downarrow$	#G $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	Time $\downarrow$	#G $\downarrow$

Appendix

Table 4: Extended evaluation metrics for all Indoor scenes. Each scene averaged from 6 independent runs.

Scene	Garden				Bicycle
	PSNR $\uparrow$	SSIM $\uparrow$	Time $\downarrow$	#G $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	Time $\downarrow$	#G $\downarrow$
MCMC 1M	28.27	0.866	7m53s	1.000	25.07	0.790	6m59s	1.000
ImprovedGS+ 1M	28.47	0.871	6m26s	1.000	25.39	0.798	5m33s	1.000
ImprovedGS+	29.39	0.896	11m39s	2.930	26.88	0.813	9m45s	3.000
ADC	28.46	0.870	13m42s	4.082	25.68	0.803	14m06s	5.211
Scene	Flowers				Treehill

Appendix

Table 5: Extended evaluation metrics for all Outdoor scenes. Each scene averaged from 6 independent runs.

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

EffGS: Efficient and High-Fidelity Gaussian Splatting
Sep 30, 2026Changbai Li, Shuo Yang, Yichen Yang +23D Gaussian Splatting 3D Gaussian Pruning
Beihang University · Nanyang Technological University
Laplacian Frequency Hierarchies for Efficient 3D Gaussian Splatting Training
Sep 3, 2026Yixiong Yang, Sisheng Zhang, Qingsong Yan +23D Gaussian Splatting 3DGS Acceleration
Harbin Institute of Technology, Shenzhen, China · XGRIDS, China
ACE-GS: Acing the Trade-off with Accurate, Compact and Efficient 3D Gaussian Splatting
Jun 19, 2026Jijian Zhao3DGS Compression 3D Gaussian Splatting
Huazhong University of Science and Technology, Wuhan, China

Sep 30, 2026J/K move · Enter open · S save

EffGS: Efficient and High-Fidelity Gaussian Splatting

Changbai Li, Shuo Yang, Yichen Yang, Shuwei Shao+1

Beihang University · Nanyang Technological University

3D Gaussian Splatting 3D Gaussian Pruning 3DGS Acceleration Adaptive Gaussian Densification Novel View Synthesis 3DGS Training

Open paper AskPDF arXiv

3D Gaussian Splatting (3DGS) enables real-time novel view synthesis, but existing general-purpose acceleration methods suffer severe rendering quality degradation when extended to more complex, large-scale scenes. To address this issue, we propose EffGS, a more general acceleration framework that improves training and rendering efficiency while maintaining reconstruction quality comparable to or better than vanilla 3DGS across bounded and large-scale scenes. EffGS combines frequency-aware guidance, localized density control, and adaptive primitive scale modulation. First, an importance scoring mechanism combines pixel-wise reconstruction errors with a difference-of-Gaussians mask scheduled over training to provide stage-dependent spatial guidance. Second, localized densification and pruning restricts density modifications to Gaussians with valid projected footprints in the sampled views. Third, learnable per-Gaussian scale modulation adjusts effective primitive extent during optimization while retaining the Compact Box rasterization rule. Extensive experiments on bounded and large-scale scene datasets demonstrate a favorable balance between reconstruction quality, training time, and primitive count. Component ablations and matched-primitive-budget comparisons further support the effectiveness of the framework.

Abstract

3D Gaussian Splatting

Explore similar work

EffGS: Efficient and High-Fidelity Gaussian Splatting

Laplacian Frequency Hierarchies for Efficient 3D Gaussian Splatting Training

ACE-GS: Acing the Trade-off with Accurate, Compact and Efficient 3D Gaussian Splatting