Reproduce “PUF: Plug-and-Play Uncertainty-Aware Fusion for Online 3D Scene Graph Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yi Yang, Myrna Castillo, Bodo Rosenhahn, Michael Ying Yang
PUF (Plug-and-Play Uncertainty-Aware Fusion) is a framework for online 3D scene graph generation that explicitly models and propagates uncertainty from 2D observations through the 3D fusion process. It achieves an 18.1-point improvement in relationship recall over the strongest online baseline, FROSS, on the 3DSSG benchmark while maintaining real-time processing.
Key insights
- Explicitly accounting for and propagating observation, 2D model, and 3D representation uncertainties through the fusion pipeline significantly improves 3D scene graph generation.
- A probabilistic approach to node association and soft evidence accumulation via Dirichlet distributions is more robust and accurate than rigid, deterministic merging strategies.
- Bayesian integration of an informative prior can effectively address the challenge of missing relational evidence for rarely observed object pairs in online incremental construction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.