Efficient Compression of Structured and Unstructured Volumes via Learned 3D Gaussian Representation

2026-07-01

AI Summary

Existing implicit neural representations struggle to encode geometric information when compressing unstructured volumetric data, often requiring additional storage of explicit meshes, which limits compression efficiency. This work proposes the first explicit volumetric compression method based on 3D Gaussian primitives, treating a collection of Gaussians as a scalar field and reconstructing point values through weighted aggregation of intersecting Gaussians. A CUDA-accelerated sampling pipeline is designed for efficient evaluation. The approach naturally embeds geometric information without relying on mesh storage and introduces a geometry-aware loss coupled with a sampling-error-driven densification strategy, significantly enhancing compression performance. Experiments demonstrate that the method achieves reconstruction quality comparable to implicit approaches on structured data with faster training, while consistently outperforming existing techniques across all metrics on unstructured data.

Abstract

Recent work has shown that implicit neural representations (INRs) can be trained to effectively compress structured and unstructured volume data, allowing for direct data querying with a reduced memory footprint. However, as existing INRs for unstructured volumes do not encode geometry, they require partial mesh storage for later sampling, limiting achievable compression. At the same time, novel view synthesis methods have shown that explicit collections of 3D Gaussians can be used to accurately visualize volume data. In this work, we introduce an explicit model for volume data compression based on 3D Gaussian primitives. We reinterpret collections of 3D Gaussians as an explicit representation of a scalar field and use a sampling strategy that reconstructs scalar values at spatial locations through weighted aggregation of intersecting Gaussians. We develop optimized CUDA-accelerated pipelines for structured and unstructured model sampling, loss functions that encourage accurate domain encoding by our models, and a novel sampling-error based densification strategy. Our explicit formulation naturally encodes domain geometry, eliminating the need for mesh storage in unstructured volumes and introducing significantly higher compression opportunities. Compared to existing INRs, we demonstrate that our explicit model achieves competitive reconstruction quality with significant training speedups on structured volumes, while markedly outperforming in all metrics on unstructured volumes.

Problem

• volume compression
• implicit neural representations
• unstructured volumes
• geometry encoding
• 3D Gaussian representation

Innovation

• 3D Gaussian representation
• explicit volume compression
• unstructured volumes
• geometry encoding
• sampling-error densification