3DTurboQuant: Training-Free Near-Optimal Quantization for 3D Reconstruction Models

Jae Joong Lee

cs.CVApr 7, 2026

3DTurboQuant: Training-Free Near-Optimal Quantization for 3D Reconstruction Models

Authors: Jae Joong Lee

Organizations: Department of Computer Science, Purdue University

Abstract

Every existing method for compressing 3D Gaussian Splatting, NeRF, or transformer-based 3D reconstructors requires learning a data-dependent codebook through per-scene fine-tuning. We show this is unnecessary. The parameter vectors that dominate storage in these models, 45-dimensional spherical harmonics in 3DGS and 1024-dimensional key-value vectors in DUSt3R, fall in a dimension range where a single random rotation transforms any input into coordinates with a known Beta distribution. This makes precomputed, data-independent Lloyd-Max quantization near-optimal, within a factor of 2.7 of the information-theoretic lower bound. We develop 3D, deriving (1) a dimension-dependent criterion that predicts which parameters can be quantized and at what bit-width before running any experiment, (2) norm-separation bounds connecting quantization MSE to rendering PSNR per scene, (3) an entry-grouping strategy extending rotation-based quantization to 2-dimensional hash grid features, and (4) a composable pruning-quantization pipeline with a closed-form compression ratio. On NeRF Synthetic, 3DTurboQuant compresses 3DGS by 3.5x with 0.02dB PSNR loss and DUSt3R KV caches by 7.9x with 39.7dB pointmap fidelity. No training, no codebook learning, no calibration data. Compression takes seconds. The code will be released (https://github.com/JaeLee18/3DTurboQuant)

Figures & tables

Bits ( $b$ )	PSNR (dB)	$\Delta$ PSNR (dB)	Compression	Quant Time	SH MSE
fp32 (baseline)	29.80	0.00	1.0 $\times$	–	–
1	29.31	$-$ 0.49	4.1 $\times$	4.2 s	0.00199
2	29.68	$-$ 0.12	3.8 $\times$	6.6 s	0.00063
3	29.78	$\boldsymbol{-}$ 0.02	3.5 $\boldsymbol{\times}$	9.3 s	0.00018
4	29.80	$-$ 0.00	3.3 $\times$	12.2 s	0.00005

Table 1: 3DTurboQuant compression of 3DGS on Lego (232K Gaussians, baseline PSNR = 29.80 dB). Rendering PSNR over 200 test views. Render time is constant (

\sim

0.8 s) as dequantization is negligible.

Panel of Figure 2 — Figure 2: Qualitative results of 3DTurboQuant on 3DGS. Rendered images across bit widths $b=1,2,3,4$ on the Lego (top) and Mic (bottom) scenes. Rightmost column: 10 $\times$ amplified error map at $b=1$ relative to the fp32 baseline. At $b=3$ , the renders are visually indistinguishable from the uncompressed model.

Panel of Figure 3 — Figure 3: DUSt3R KV cache quantization: depth map visualization. Predicted depth maps (turbo colormap) from DUSt3R ViT-Large with KV cache quantized at various bit widths. At $b=4$ (7.9 $\times$ KV compression, 39.7 dB pointmap PSNR), the depth structure is indistinguishable from the unquantized baseline.

Configuration	Gaussians	PSNR (dB)	$\Delta$ PSNR (dB)	Ratio
TQ $b\!=\!3$ (quant only)	232,743	29.78	$-$ 0.02	3.5 $\times$
TQ $b\!=\!3$ + prune $\tau\!=\!0.05$	196,887	29.63	$-$ 0.17	4.1 $\times$
TQ $b\!=\!3$ + prune $\tau\!=\!0.1$	173,482	28.98	$-$ 0.82	4.7 $\times$
TQ $b\!=\!3$ + prune $\tau\!=\!0.2$	144,022	27.21	$-$ 2.59	5.6 $\times$
TQ $b\!=\!3$ + SH $1$	232,743	28.06	$-$ 1.74	4.3 $\times$
TQ $b\!=\!3$ + prune $\tau\!=\!0.3$ + SH $1$	123,863	25.05	$-$ 4.75	8.0 $\times$

Table 2: Pruning + 3DTurboQuant on Lego. All configurations are training-free.

\tau

: opacity threshold. SH

l^{\prime}

: reduced SH degree.

Bits ( $b$ )	Ptmap PSNR (dB)	3D Point MSE	KV Compress	Inf. Time	Overhead
fp32	$\infty$	0	1.0 $\times$	0.14 s	–
1	16.52	0.01386	31.0 $\times$	1.04 s	+0.90 s
2	16.52	0.01386	15.8 $\times$	1.85 s	+1.72 s
3	29.30	0.00078	10.6 $\times$	0.94 s	+0.81 s
4	39.68	0.00007	7.9 $\boldsymbol{\times}$	1.67 s	+1.53 s
5	49.65	0.000008	6.4 $\times$	2.32 s	+2.19 s

Table 3: 3DTurboQuant KV cache quantization in DUSt3R ViT-Large (571M params, 48 attention layers,

d_{\text{kv}}\!=\!1024

). Baseline inference: 0.14 s. Overhead = additional time from quantization.

Bits ( $b$ )	PSNR (dB)	$\Delta$ PSNR (dB)	Hash Compress	Quant Time
fp32	11.57	0.00	1.0 $\times$	–
1	9.70	$-$ 1.87	1.9 $\times$	0.18 s
2	10.54	$-$ 1.04	1.8 $\times$	0.23 s
4	10.51	$\boldsymbol{-}$ 1.07	1.6 $\boldsymbol{\times}$	0.91 s
8	10.49	$-$ 1.08	1.3 $\times$	11.5 s

Table 4: 3DTurboQuant on Instant-NGP hash features (Lego). Low ratios reflect the 2D per-entry feature dimension; higher-dim representations (TensoRF

d_{f}\!=\!48

, K-Planes

d_{f}\!=\!64

) would yield 3–7

\times

Training-required methods
Method	Venue	Compress	PSNR Loss	Training	Time
LightGaussian Fan et al. (2024)	NeurIPS’24	15 $\times$	0.2–0.5 dB	Yes	Hours
ContextGS Wang et al. (2024b)	NeurIPS’24	20 $\times$	0.1–0.3 dB	Yes	Hours
C3DGS Niedermayr et al. (2024)	CVPR’24	31 $\times$	0.1–0.5 dB	Yes	Hours
SOGS Morgenstern et al. (2024)	ECCV’24	17–42 $\times$	0.1–0.5 dB	Yes	Hours
FCGS Chen et al. (2025a)	ICLR’25	$>$ 20 $\times$	$\sim$ 0.1 dB	Yes	Seconds

Table 5: Comparison with existing 3DGS compression methods. Prior methods combine pruning, learned codebooks, entropy coding, and per-scene fine-tuning. 3DTurboQuant provides provably near-optimal quantization only, with no training. Compression ratios are reported relative to vanilla 3DGS.

3DGS Compression

3D Gaussian Splatting compression reduces storage and rendering costs while preserving the quality of reconstructed views and scenes.

See all 85 papers

MesonGS++: Post-training Compression of 3D Gaussian Splatting with Hyperparameter Searching
Apr 29, 2026Shuzhao Xie, Junchen Ge, Weixiang Zhang +103DGS Compression 3D Gaussian Splatting
Shenzhen International Graduate School, Tsinghua University, Shenzhen, China · The Hong Kong University of Science and Technology, Hong Kong SAR, China · Harbin Institute of Technology, Shenzhen, China +6
GETA-3DGS: Automatic Joint Structured Pruning and Quantization for 3D Gaussian Splatting
May 3, 2026Baobing Zhang, Wanxin Sui3DGS Compression Mixed-Precision Quantization
Robotics Research Group, School of Physics, Engineering and Computer Science, University of Hertfordshire, Hatfield, AL10 9AB, United Kingdom · Department of Electronic and Electrical Engineering, Brunel University London, Kingston Lane, Uxbridge, Middlesex, UB8 3PH, United Kingdom
Squeeze3D: Extreme Neural Compression with Latent Space Bridging
Jun 9, 2025Rishit Dagli, Yushi Guan, Sankeerth Durvasula +2Point Cloud Compression
University of Toronto

Apr 29, 2026J/K move · Enter open · S save

MesonGS++: Post-training Compression of 3D Gaussian Splatting with Hyperparameter Searching

Shuzhao Xie, Junchen Ge, Weixiang Zhang, Jiahang Liu+9

Shenzhen International Graduate School, Tsinghua University, Shenzhen, China · The Hong Kong University of Science and Technology, Hong Kong SAR, China · Harbin Institute of Technology, Shenzhen, China+6

3DGS Compression 3D Gaussian Splatting 3D Gaussian Pruning Post-Training Quantization Gaussian Splatting Compression

Open paper AskPDF arXiv

3D Gaussian Splatting (3DGS) achieves high-quality novel view synthesis with real-time rendering, but its storage cost remains prohibitive for practical deployment. Existing post-training compression methods still rely on many coupled hyperparameters across pruning, transformation, quantization, and entropy coding, making it difficult to control the final compressed size and fully exploit the rate-distortion trade-off. We propose MesonGS++, a size-aware post-training codec for 3D Gaussian compression. On the codec side, MesonGS++ combines joint importance-based pruning, octree geometry coding, attribute transformation, selective vector quantization for higher-degree spherical harmonics, and group-wise mixed-precision quantization with entropy coding. On the configuration side, it treats the reserve ratio and bit-width allocation as the dominant rate-distortion knobs and jointly optimizes them under a target storage budget via discrete sampling and 0--1 integer linear programming. We further propose a linear size estimator and a CUDA parallel quantization operator to accelerate the hyperparameter searching process. Extensive experiments show that MesonGS++ achieves over 34

\times

compression while preserving rendering fidelity, outperforming state-of-the-art post-training methods and accurately meeting target size budgets. Remarkably, without any training, MesonGS++ can even surpass the PSNR of vanilla 3DGS at a 20

\times

compression rate on the Stump scene. Our code is available at https://github.com/mmlab-sigs/mesongs_plus

Abstract

3DGS Compression

Explore similar work

MesonGS++: Post-training Compression of 3D Gaussian Splatting with Hyperparameter Searching

GETA-3DGS: Automatic Joint Structured Pruning and Quantization for 3D Gaussian Splatting

Squeeze3D: Extreme Neural Compression with Latent Space Bridging