3D Gaussian Splatting (3DGS) has emerged as a leading technology for high-quality 3D scene reconstruction. However, the iterative refinement and densification process leads to the generation of a large number of primitives, each contributing to the reconstruction to a substantially different extent. Estimating primitive importance is thus crucial, both for removing redundancy during reconstruction and for enabling efficient compression and transmission. Existing methods typically rely on rendering-based analyses, where each primitive is evaluated through its contribution across multiple camera viewpoints. However, such methods are sensitive to the number and selection of views, rely on specialized differentiable rasterizers, and have long calculation times that grow linearly with view count, making them difficult to integrate as plug-and-play modules and limiting scalability and generalization. To address these issues, we propose RAP, a fast feedforward rendering-free attribute-guided method for efficient importance score prediction in 3DGS. RAP infers primitive significance directly from intrinsic Gaussian attributes and local neighborhood statistics, avoiding rendering-based or visibility-dependent computations. A compact MLP predicts per-primitive importance scores using rendering loss, pruning-aware loss, and significance distribution regularization. After training on a small set of scenes, RAP generalizes effectively to unseen data and can be seamlessly integrated into reconstruction, compression, and transmission pipelines. Our code is publicly available at https://github.com/yyyykf/RAP.
Figures & tables
Figure 1: Visualization of redundant and low-importance Gaussians in the Train scene from Tanks&Temples.
Figure 2: RAP learning framework.
Figure 3: PSNR vs. retention ratio curves.
LightGS
MesonGS
EAGLES
C3DGS
PUP-3DGS
RAP
Mip-Outdoor
-35.21
-34.89
-41.28
-33.77
-22.54
-42.63
Mip-Indoor
-31.15
-30.34
-24.98
-27.63
-8.70
-33.90
Deep Blending
-30.72
-28.84
-29.87
-16.28
-19.46
-36.76
Tanks&Temples
-37.98
-36.12
-30.01
-34.37
7.06
-40.11
Table 1: BD-Rate (%) comparison of pruning-based Gaussian Splatting methods relative to the opacity baseline. Lower BD-Rate indicates better rate–distortion efficiency
Dataset
Opacity
LightGS
MesonGS
EAGLES
C3DGS
PUP-3DGS
RAP
Mip-Outdoor
3.73
15.86
42.40
42.96
11.69
20.95
15.64
Mip-Indoor
1.27
22.71
14.84
14.97
15.96
21.22
5.72
Deep Blending
2.99
20.20
31.30
31.65
13.62
20.28
11.64
Tanks&Temples
2.53
18.62
17.53
13.92
9.22
20.50
6.66
Table 2: Importance score computation time (seconds).
Method
MipNeRF360 Outdoor
MipNeRF360 Indoor
Deep Blending
Tanks & Temples
PSNR ↑
SSIM ↑
LPIPS ↓
Size ↓
PSNR ↑
SSIM ↑
LPIPS ↓
Size ↓
PSNR ↑
SSIM ↑
LPIPS ↓
Size ↓
PSNR ↑
SSIM ↑
LPIPS ↓
Size ↓
3DGS
24.688
0.729
0.238
926.895
31.076
0.925
0.186
299.202
29.817
0.907
0.239
586.064
23.734
0.853
0.169
372.036
Opacity
24.550
0.718
0.275
313.657
30.604
0.912
0.216
72.176
29.899
0.910
0.248
149.873
23.674
0.842
0.200
118.414
LightGaussian
24.676
0.728
0.255
300.907
30.777
0.917
0.205
71.676
29.839
0.908
0.249
155.207
23.634
0.845
0.191
118.048
MesonGS
24.554
0.723
0.256
297.255
30.173
0.914
0.207
71.587
29.638
0.904
0.254
154.137
23.541
0.839
0.197
115.922
EAGLES
24.520
0.722
0.256
295.899
29.801
0.911
0.210
71.356
29.551
0.904
0.254
154.352
23.212
0.834
0.202
113.500
Table 3: Reconstruction quality and size comparison under integrated pruning (40% pruning every 1500 iterations).
Figure 4: R-D curves for GSC with 20% pruning
Figure 5: Ablation studies of the proposed feature and loss.
Figure 6: Robustness of rendering-based importance estimation to the number of training views used. Left: LightGaussian on the bicycle scene from Mip-NeRF360-Outdoor. Right: C3DGS on the playroom scene from Deep Blending. Each curve corresponds to a different subset of views ( 1/8 – 7/8 ) used for score computation.
Figure 7: Pruning behavior of different importance estimators at 5% retention. EAGLES favors central regions, C3DGS keeps edges, while RAP produces a more uniform, structure-preserving subset.
Figure 8: Predicted score distributions for RAP, RAP without pruning-aware loss, and RAP without distribution loss.
Figure 9: Per-scene post-hoc pruning results across three metrics: PSNR (top), SSIM (middle), and LPIPS (bottom).
Existing pruning methods for 3D Gaussian splatting (3DGS) suffer from either severe quality degradation or prohibitive computational overhead. In this paper, we propose REFINE, a highly accelerated 3DGS pruning framework centered on a novel rendering-free primitive importance metric. Our approach leverages an analytically approximated, rendering-aware Hessian field to quantify the expected perceptual error induced by the removal of individual primitives. By modeling the joint modulation of visibility, projection geometry and the content adaptive hyperparameter, we entirely bypass costly forward rendering passes and derive an anisotropic perceptual weight field that serves as a high-fidelity proxy for primitive importance. Extensive experiments across multiple benchmark datasets demonstrate that REFINE maintains highly competitive rendering quality while achieving a 3,000× reduction in pruning-related computational complexity, translating to a practical ∼20× speedup in device latency compared to state-of-the-art pruning methods.
Zhang Chen, Shuai Wan, Mengting Yu +2
School of Electronics and Information, Northwestern Polytechnical University · School of Telecommunication Engineering, Xidian University · Department of Computer Science, City University of Hong Kong
Recent significant advances in 3D scene representation have been driven by 3D Gaussian Splatting (3DGS), which has enabled real-time rendering with photorealistic quality. 3DGS often requires a large number of primitives to achieve high fidelity, leading to redundant representations and high resource consumption, thereby limiting its scalability for complex or large-scale scenes. Consequently, effective pruning strategies and more expressive primitives that can reduce redundancy while preserving visual quality are crucial for practical deployment. We propose an efficient, integrated reconstruction-aware pruning strategy that adaptively determines pruning timing and refining intervals based on reconstruction quality, thus reducing model size while enhancing rendering quality. Moreover, we introduce a 3D Difference-of-Gaussians primitive that jointly models both positive and negative densities in a single primitive, improving the expressiveness of Gaussians under compact configurations. Our method significantly improves model compactness, achieving up to 90% reduction in Gaussian-count while delivering visual quality that is similar to, or in some cases better than, that produced by state-of-the-art methods. Code will be made publicly available.
Haoran Wang, Guoxi Huang, Fan Zhang +2
School of Computer Science, University of Bristol, Bristol, UK
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.
Changbai Li, Shuo Yang, Yichen Yang +2
Beihang University · Nanyang Technological University