Radiant: Large-scale 3D Gaussian Rendering based on Hierarchical Framework

TL;DR

Radiant is a hierarchical 3D Gaussian Splatting framework that improves large-scale scene reconstruction by considering system heterogeneity, optimizing workload distribution, and enhancing model quality and efficiency.

Contribution

It introduces a novel hierarchical 3DGS algorithm that accounts for system heterogeneity and workload partitioning, enabling efficient large-scale scene reconstruction.

Findings

Reconstruction quality improved by up to 25.7%.

End-to-end latency reduced by up to 79.6%.

Effective workload partitioning and camera allocation are crucial.

Abstract

With the advancement of computer vision, the recently emerged 3D Gaussian Splatting (3DGS) has increasingly become a popular scene reconstruction algorithm due to its outstanding performance. Distributed 3DGS can efficiently utilize edge devices to directly train on the collected images, thereby offloading computational demands and enhancing efficiency. However, traditional distributed frameworks often overlook computational and communication challenges in real-world environments, hindering large-scale deployment and potentially posing privacy risks. In this paper, we propose Radiant, a hierarchical 3DGS algorithm designed for large-scale scene reconstruction that considers system heterogeneity, enhancing the model performance and training efficiency. Via extensive empirical study, we find that it is crucial to partition the regions for each edge appropriately and allocate varying camera positions to each device for image collection and training. The core of Radiant is partitioning regions based on heterogeneous environment information and allocating workloads to each device accordingly. Furthermore, we provide a 3DGS model aggregation algorithm that enhances the quality and ensures the continuity of models' boundaries. Finally, we develop a testbed, and experiments demonstrate that Radiant improved reconstruction quality by up to 25.7\% and reduced up to 79.6\% end-to-end latency.