Reproduce “Paparazzo: Active Mapping of Moving 3D Objects” 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
Davide Allegro, Shiyao Li, Stefano Ghidoni, Vincent Lepetit
Researchers from the University of Padova and LIGM introduce Paparazzo, a learning-free framework for actively reconstructing independently moving 3D objects using 3D Gaussian Splatting and an Extended Kalman Filter for motion prediction. This system systematically selects viewpoints, accounting for both information gain and temporal synchronization, achieving up to 10% higher 3D coverage and improved completeness compared to passive tracking baselines across diverse dynamic scenarios.
Key insights
- Active viewpoint selection for dynamic objects must consider both the geometric informativeness of a pose and the temporal synchronization required to reach that pose as the object moves.
- An adaptive control strategy, switching between a tracking-focused mode and a mapping-focused mode based on motion estimation confidence, enhances robustness to complex and unpredictable object motions.
- Leveraging an EKF for continuous motion prediction allows the mapping agent to anticipate future object states, enabling more effective path planning for dynamic reconstruction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.