DELUGE: Decomposed Entropy-coded Live Unstructured Geometry Exchange for Real-time Particle Streaming

arXiv cs.GR details DELUGE, a new compression architecture for streaming particle-based simulations to clients in real time. The work is aimed at VR and AR scenarios where multiple users need to see the same fluid, smoke, or granular simulation without the usual latency spikes that make point-cloud codecs struggle.

The core idea is to exploit temporal coherence and predictable particle motion instead of treating the data like a static point cloud. That matters because common codecs such as G-PCC (TMC13) and Draco can be too slow to encode for frame-by-frame delivery when the geometry is changing every frame.

The authors report roughly 20× faster decoding than G-PCC and 6× faster encoding than Draco on dynamic point-cloud datasets. They also built end-to-end clients for web browsers and Apple Vision Pro, then tested the system in perceptual and two-person collaborative tasks.

For developers building social VR, digital twins, or networked simulation tools, the practical takeaway is that particle-heavy effects may be streamable with much lower latency than traditional geometry pipelines allow. The exact production constraints still depend on bandwidth, scene complexity, and how much simulation state must be synchronized alongside the particles.