S2Tok: Streaming 3D Gaussian Reconstruction with Persistent Spatial Tokens

Key points

• Streaming 3D reconstruction requires more than a sequence of geometric predictions: it requires a persistent scene state that can incorporate new evidence and remain renderable as observations arrive.1
• A hierarchical decoder and Gaussian head convert the evolving state into non-pixel-aligned 3D Gaussians, enabling novel-view rendering without caching previous frames.1
• Experiments across four benchmarks demonstrate competitive streaming rendering quality with compact Gaussian representations.1
• These results support latent spatial tokens as a persistent computational state for online 3D reconstruction, combining learned scene updates with explicit Gaussian rendering.1

Sources (1)

• [1]S2Tok: Streaming 3D Gaussian Reconstruction with Persistent Spatial Tokens
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 06:40 PM

  We introduce S2Tok, a feed-forward framework that maintains a size-adaptive, persistent scene state from uncalibrated image streams.

  Streaming 3D reconstruction requires more than a sequence of geometric predictions: it requires a persistent scene state that can incorporate new evidence and remain renderable as observations arrive.

Extractive summary: sentences quoted from the sources.