Search papers, labs, and topics across Lattice.
This paper introduces LagrangeGS, a novel formulation of dynamic 3D Gaussian Splatting (3DGS) as a non-conservative Lagrangian system to address key limitations in existing physics-aware extensions. By approximating the velocity-Hessian as an identity matrix and restricting non-conservative forces to be time-independent, LagrangeGS achieves stable long-term extrapolation and consistent time-reversibility in dynamic scene reconstructions. Extensive evaluations show that this approach not only resolves issues of physically inconsistent trajectories and geometric collapse but also facilitates counterfactual physics-based editing without requiring retraining.
LagrangeGS enables stable long-term extrapolation and time-reversibility in dynamic scene reconstruction, overcoming critical limitations of existing methods.
Dynamic 3D Gaussian Splatting (3DGS) achieves photorealistic reconstruction of time-varying scenes, and recent physics-aware extensions improve extrapolation by explicitly predicting velocity fields. However, these extensions merely fit vector fields to visual deformations without satisfying Lagrangian mechanics, leading to three major issues: (i) physically inconsistent trajectories, (ii) lack of time-reversibility, and (iii) geometric collapse during long-term extrapolation. In this paper, we propose LagrangeGS, which formulates dynamic 3DGS as a non-conservative Lagrangian system. While this Lagrangian formulation fundamentally solves (i), a direct application of general LNNs to dynamic 3DGS requires a large velocity-Hessian inversion for millions of Gaussian particles. To overcome this computational bottleneck, we approximate the velocity-Hessian as an identity matrix, decoupling particle dynamics for computational tractability. For (ii), we restrict the non-conservative forces to be explicitly time independent, enabling consistent backward integration. Finally, to address (iii), we introduce local rigid alignment that regularizes particle trajectories. Extensive evaluations on dynamic scene benchmarks demonstrate that LagrangeGS enables stable long-term extrapolation, consistent time reversal, and counterfactual physics-based editing without retraining.
