3D Gaussian Splatting for Scientific Particle Data Compression and Rendering

2026-07-24

AI Summary

This work addresses the high storage, transmission, and interactive visualization costs of large-scale particle simulations, where existing compression methods often fail to preserve downstream rendering fidelity. The authors propose ParticleGS, a novel framework that introduces 3D Gaussian splatting into scientific particle data compression. By leveraging multi-stage multi-view training, a lightweight VizMapper network, and KD-tree-based spatial partitioning, ParticleGS learns a compact representation explicitly optimized for visual quality. The method dynamically adapts to varying rendering parameters at inference time without requiring retraining. Evaluated on a 281-million-particle HACC simulation, ParticleGS achieves a 65× compression ratio with a PSNR of 30.03 dB—5–8 dB higher than SZ3—while enabling real-time rendering at 662 FPS on a single GPU, over 2,300× faster than ParaView, and demonstrates strong generalization across diverse simulation datasets.

Abstract

Large-scale particle simulations produce hundreds of millions of particles, straining storage, transfer, and interactive visualization. Existing lossy compressors such as SZ3 operate in data space and provide no guarantees on downstream visualization fidelity. We propose ParticleGS, a visualization-aware framework based on 3D Gaussian Splatting (3DGS) that learns a compact representation directly optimized for rendered image quality, combining (1) a multi-stage, multi-orbit training pipeline, (2) VizMapper, a lightweight network that adapts a single trained model to user-specified visualization parameters at inference time, and (3) spatial block training with KD-tree decomposition and global fine-tuning. On a 281-million-particle HACC cosmological simulation, our 8-block model reaches 30.03 dB PSNR at 65x compression, outperforming SZ3 by 5-8 dB at comparable ratios, and generalizes without tuning to additional HACC regions and a dark-matter-only FIRE-2 simulation. It renders at 662 FPS on a single GPU, over 2,300x faster than ParaView on the full particle data.

Problem

• particle data compression
• visualization fidelity
• 3D Gaussian Splatting
• scientific visualization
• large-scale simulation

Innovation

• 3D Gaussian Splatting
• visualization-aware compression
• VizMapper
• KD-tree decomposition
• particle data rendering