Search papers, labs, and topics across Lattice.
This paper introduces the Spatiotemporal Position Implicit Network for 4D Gaussian Splatting (SPIN-4DGS), which addresses the limitations of existing methods in reconstructing fast-moving objects with large inter-frame displacements. By leveraging explicitly collected spatiotemporal positions to learn Gaussian attributes, SPIN-4DGS enhances the fidelity of splatting under challenging motion conditions. The method demonstrates significant improvements in image quality metrics, achieving a +1.83 increase in PSNR over the leading baseline on complex sports scenes from the CMU Panoptic dataset.
SPIN-4DGS achieves unprecedented fidelity in Gaussian splatting for fast motion, outperforming existing methods by a notable margin.
Recent 4D Gaussian Splatting (4DGS) methods often fail under fast motion with large inter-frame displacements, where Gaussian attributes are poorly learned during training, and fast-moving objects are often lost from the reconstruction. In this work, we introduce Spatiotemporal Position Implicit Network for 4DGS, coined SPIN-4DGS, which learns Gaussian attributes from explicitly collected spatiotemporal positions rather than modeling temporal displacements, thereby enabling more faithful splatting under fast motions with large inter-frame displacements. To avoid the heavy memory overhead of explicitly optimizing attributes across all spatiotemporal positions, we instead predict them with a lightweight feed-forward network trained under a rasterization-based reconstruction loss. Consequently, SPIN-4DGS learns shared representations across Gaussians, effectively capturing spatiotemporal consistency and enabling stable high-quality Gaussian splatting even under challenging motions. Across extensive experiments, SPIN-4DGS consistently achieves higher fidelity under large displacements, with clear improvements in PSNR and SSIM on challenging sports scenes from the CMU Panoptic dataset. For example, SPIN-4DGS notably outperforms the strongest baseline, D3DGS, by achieving +1.83 higher PSNR on the Basketball scene.
