Search papers, labs, and topics across Lattice.
The paper introduces TRiGS, a 4D Gaussian Splatting method that models dynamic scenes using continuous, geometrically consistent rigid-body transformations based on $SE(3)$ transformations, hierarchical Bezier residuals, and learnable local anchors. This approach addresses the temporal fragmentation issues of existing 4DGS methods that rely on piecewise linear velocity approximations, leading to primitive regeneration and memory growth. TRiGS achieves state-of-the-art rendering fidelity and, crucially, scales to significantly longer video sequences (600-1200 frames) with improved temporal stability and reduced memory consumption compared to prior art.
Forget memory bottlenecks: TRiGS lets 4D Gaussian Splatting handle 10x longer videos by encoding temporal dynamics with continuous rigid-body transformations.
Recent 4D Gaussian Splatting (4DGS) methods achieve impressive dynamic scene reconstruction but often rely on piecewise linear velocity approximations and short temporal windows. This disjointed modeling leads to severe temporal fragmentation, forcing primitives to be repeatedly eliminated and regenerated to track complex nonlinear dynamics. This makeshift approximation eliminates the long-term temporal identity of objects and causes an inevitable proliferation of Gaussians, hindering scalability to extended video sequences. To address this, we propose TRiGS, a novel 4D representation that utilizes unified, continuous geometric transformations. By integrating $SE(3)$ transformations, hierarchical Bezier residuals, and learnable local anchors, TRiGS models geometrically consistent rigid motions for individual primitives. This continuous formulation preserves temporal identity and effectively mitigates unbounded memory growth. Extensive experiments demonstrate that TRiGS achieves high fidelity rendering on standard benchmarks while uniquely scaling to extended video sequences (e.g., 600 to 1200 frames) without severe memory bottlenecks, significantly outperforming prior works in temporal stability.
