Volumetric capture has a distribution problem. Capturing a performer as a live point cloud is now routine — depth cameras are cheap, and the software is mature. Getting that cloud to somebody else’s headset, live, is where projects die.
A video stream is a grid of pixels, in a fixed order, changing predictably. Codecs have had thirty years of work poured into exploiting that. A point cloud is none of those things: an unordered pile of positions and colours, different in count and arrangement every frame, with no notion of “the pixel to the left.” Almost everything that makes video compression efficient is unavailable.
DELUGE — Decomposed Entropy-coded Live Unstructured Geometry Exchange for Real-time Particle Streaming — accepted to IEEE ISMAR 2026 and published in TVCG, is aimed squarely at this.
What the name tells you
The title is unusually descriptive, and it’s worth unpacking because it lays out the approach:
- Decomposed — the geometry is split into components rather than compressed as one blob. Position, colour and other attributes have very different statistical behaviour, and separating them lets each be coded appropriately.
- Entropy-coded — after decomposition, each stream is compressed according to how predictable it actually is. Entropy coding is the mathematically principled way of spending fewer bits on the parts that surprise you less.
- Live and real-time — the constraint that rules out most of the existing point-cloud compression literature, which assumes you can take your time.
- Unstructured geometry — no mesh, no fixed topology, no frame-to-frame correspondence.
Why this beat cares
Three kinds of work are blocked on exactly this problem:
- Volumetric telepresence and remote performance. A dancer captured in one city appearing as a point cloud in headsets in another is a well-established artistic idea whose implementations are almost always pre-recorded, because live transmission doesn’t hold up.
- Multi-site installations that share a live captured space between venues.
- Anything involving a captured person rather than an avatar, which matters because point clouds carry presence in a way rigged avatars don’t — the noise, the holes and the shimmer read as a real body imperfectly seen, and audiences respond to that differently.
It’s an engineering paper, not a creative one, and its home venue is immersive visualisation rather than art. But the gap between “volumetric capture works” and “volumetric capture works between two rooms in different countries” is pure transport, and that gap is what has kept live volumetric performance an expensive specialist undertaking rather than something a small company can tour.
Worth watching whether the code appears — like most of this field, adoption will depend entirely on whether someone wraps it in something a TouchDesigner or Unreal user can call.
Related Reading
- DELUGE: Decomposed Entropy-coded Live Unstructured Geometry Exchange for Real-time Particle Streaming — arXiv:2609.19750
- IEEE Computer Society — conferences
- IEEE Transactions on Visualization and Computer Graphics
- MPEG point cloud compression (V-PCC/G-PCC) — background
- Depthkit — volumetric capture tooling used in art and performance