3DGStream: On-the-Fly Training of 3D Gaussians for Efficient Streaming of Photo-Realistic Free-Viewpoint Videos

TL;DR

3DGStream enables real-time, on-the-fly photo-realistic free-viewpoint video streaming of dynamic scenes using 3D Gaussians and neural transformation caching, significantly reducing training time and enabling fast rendering.

Contribution

We introduce 3DGStream, a novel method that achieves efficient, real-time free-viewpoint video streaming by using 3D Gaussians and a neural transformation cache for dynamic scene reconstruction.

Findings

Per-frame reconstruction within 12 seconds

Real-time rendering at 200 FPS

Competitive performance in speed and quality

Abstract

Constructing photo-realistic Free-Viewpoint Videos (FVVs) of dynamic scenes from multi-view videos remains a challenging endeavor. Despite the remarkable advancements achieved by current neural rendering techniques, these methods generally require complete video sequences for offline training and are not capable of real-time rendering. To address these constraints, we introduce 3DGStream, a method designed for efficient FVV streaming of real-world dynamic scenes. Our method achieves fast on-the-fly per-frame reconstruction within 12 seconds and real-time rendering at 200 FPS. Specifically, we utilize 3D Gaussians (3DGs) to represent the scene. Instead of the na\"ive approach of directly optimizing 3DGs per-frame, we employ a compact Neural Transformation Cache (NTC) to model the translations and rotations of 3DGs, markedly reducing the training time and storage required for each FVV frame. Furthermore, we propose an adaptive 3DG addition strategy to handle emerging objects in dynamic scenes. Experiments demonstrate that 3DGStream achieves competitive performance in terms of rendering speed, image quality, training time, and model storage when compared with state-of-the-art methods.