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

2026-04-06

AI Summary

This work addresses the high computational cost and complexity of existing compression methods for 3D reconstruction models—such as 3D Gaussian Splatting (3DGS), NeRF, and DUSt3R—which typically require per-scene fine-tuning to learn data-dependent codebooks. The authors propose a near-optimal, training-free quantization approach that eliminates the need for codebooks or calibration data. Leveraging the observation that randomly rotated parameter vectors follow a Beta distribution in specific dimensions, the method combines this statistical property with precomputed Lloyd-Max quantization for efficient compression. Key contributions include a dimension-aware quantization criterion, a theoretical bound linking quantization error to rendering quality, a 2D hash-grid-based feature grouping strategy, and a composable pruning-quantization pipeline. Experiments demonstrate that on the NeRF Synthetic benchmark, 3DGS achieves 3.5× compression with only 0.02 dB PSNR loss, while DUSt3R’s KV cache is compressed 7.9× with 39.7 dB pointmap fidelity, all within seconds of compression time.

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)

Problem

• 3D reconstruction
• model compression
• quantization
• training-free
• codebook learning

Innovation

• training-free quantization
• 3D reconstruction compression
• random rotation
• Lloyd-Max quantization
• parameter vector distribution