CompGS: Efficient 3D Scene Representation via Compressed Gaussian Splatting

TL;DR

CompGS tackles the large data footprint of Gaussian splatting by introducing a hybrid primitive structure and rate-constrained optimization to compress 3D scenes without sacrificing rendering quality. It uses a small set of anchor primitives to predict a larger set of coupled primitives, with coupled attributes stored as compact residual embeddings and learned warps for geometry. The optimization jointly minimizes rendering distortion and bitrate using an entropy-based bitrate model, producing an RD trade-off L = $\lambda R + D$ over all primitives. Empirical results on Tanks&Templates, Deep Blending, and Mip-NeRF 360 show dramatic compression ratios (up to $110.45\times$) with minimal PSNR loss, validating CompGS's practicality for real-world view synthesis.

Abstract

Gaussian splatting, renowned for its exceptional rendering quality and efficiency, has emerged as a prominent technique in 3D scene representation. However, the substantial data volume of Gaussian splatting impedes its practical utility in real-world applications. Herein, we propose an efficient 3D scene representation, named Compressed Gaussian Splatting (CompGS), which harnesses compact Gaussian primitives for faithful 3D scene modeling with a remarkably reduced data size. To ensure the compactness of Gaussian primitives, we devise a hybrid primitive structure that captures predictive relationships between each other. Then, we exploit a small set of anchor primitives for prediction, allowing the majority of primitives to be encapsulated into highly compact residual forms. Moreover, we develop a rate-constrained optimization scheme to eliminate redundancies within such hybrid primitives, steering our CompGS towards an optimal trade-off between bitrate consumption and representation efficacy. Experimental results show that the proposed CompGS significantly outperforms existing methods, achieving superior compactness in 3D scene representation without compromising model accuracy and rendering quality. Our code will be released on GitHub for further research.

Figures (8)

• Figure 1: Comparison between the proposed method and concurrent Gaussian splatting compression methods navaneet2023compact3dniedermayr2023compressedlee2023compactgirish2023eagles on the Tanks&Templates dataset knapitsch2017tanks. Comparison metrics include rendering quality in terms of PSNR, model size and bits per primitive.
• Figure 2: Illustration of local similarities of 3D Gaussians. The local similarity is measured by the average cosine distances between a 3D Gaussian and its 20 neighbors with minimal Euclidean distance.
• Figure 3: Overview of our proposed method.
• Figure 4: Illustration of the proposed inter-primitive prediction.
• Figure 5: Illustration of the proposed entropy estimation.