Real-Time Physics Simulation with Dynamic Mesh-Gaussian Reconstructions
For researchers in dynamic 3D reconstruction and physics simulation, this paper provides a negative result showing that post-processing topology conversion cannot reconcile high-quality reconstruction with physics-compatible topology.
The paper investigates whether dynamic 3D reconstructions with varying topology can be converted to fixed-topology meshes for physics simulation. They propose a dual-representation framework achieving 4.65× speedup, but find that conversion strategies incur 65-80% geometric degradation, concluding that reconstruction quality and physics-compatible topology are fundamentally distinct objectives.
Integrating dynamic 3D reconstructions into physics simulation requires fixed mesh topology for efficient collision detection, but state-of-the-art methods like DG-Mesh produce varying topology optimized for geometric quality. We investigate whether topology conversion can enable physics integration while preserving reconstruction fidelity. We propose a dual-representation framework combining fixed-topology meshes for physics with Gaussian splatting for rendering, achieving 4.65$\times$ speedup over varying-topology baselines through runtime vertex buffer updates. We evaluate two conversion strategies, temporal correspondence tracking and template-based projection, against native fixed-topology methods (MaGS) on the DG-Mesh dataset. Our evaluation reveals that both conversion approaches incur 65-80% geometric degradation, producing results inferior to MaGS despite DG-Mesh's superior initial quality. This demonstrates that high-quality reconstruction and physics-compatible topology represent fundamentally distinct objectives that cannot be reconciled through post-processing. Our findings inform future development of physics-aware reconstruction methods and our framework enables real-time simulation with any fixed-topology approach.