PersistGS: 4D高斯溅射中物体持久性的可微物理
PersistGS: Differentiable Physics for Object Permanence in 4D Gaussian Splatting
AI总结：
提出PersistGS方法，通过将可微刚体模拟与3D高斯溅射耦合，在物体被遮挡期间利用物理规律预测其SE(3)轨迹，从而恢复物体持久性，并引入质心轮廓损失降低轨迹误差。
摘要：
动态3D高斯溅射（3DGS）方法通过光度监督从同步多相机视频重建时变场景。当一个运动物体被所有训练相机完全遮挡时，光度监督消失：表示该物体的高斯体无法接收梯度信号而退化。现有处理神经重建中不完整观测的方法依赖于学习到的生成先验，这些先验优先考虑视觉合理性而非物理正确性。我们提出$ extbf{PersistGS}$，一种通过将可微刚体模拟与3D高斯溅射耦合来在遮挡期间恢复物体持久性的方法。我们的方法将场景分解为每个物体的高斯体和碰撞网格，通过可微模拟从观测到的遮挡前轨迹估计摩擦和速度，并利用得到的SE(3)轨迹在整个遮挡期间定位物体高斯体。由于预测轨迹满足刚体动力学的控制方程，它能够忠实捕捉接触事件（弹跳、基于摩擦的减速、方向变化），而运动学外推无法建模这些事件。我们引入质心轮廓损失，将位置梯度与外观噪声分离，使轨迹误差比光度监督降低40%。我们使用在训练中保留的相机进行评估，这些相机在遮挡期间观察物体。在合成场景上的实验表明，PersistGS在PSNR上比恒定速度外推高出2.46dB，并且与真实轨迹上限仅差0.19dB。
英文摘要：
Dynamic 3D Gaussian Splatting (3DGS) methods reconstruct time-varying scenes from synchronized multi-camera video using photometric supervision. When a moving object becomes fully occluded from all training cameras, this supervision vanishes: the Gaussians representing it receive no gradient signal and degrade. Existing approaches to incomplete observations in neural reconstruction rely on learned generative priors that prioritize visual plausibility over physical correctness. We propose $\textbf{PersistGS}$, a method that restores object permanence during occlusion by coupling differentiable rigid body simulation with 3D Gaussian Splatting. Our approach decomposes the scene into per-object Gaussians and collision meshes, estimates friction and velocity from the observed pre-occlusion trajectory via differentiable simulation, and uses the resulting SE(3) trajectory to position object Gaussians throughout the occlusion period. Because the predicted trajectory satisfies the governing equations of rigid body dynamics, it faithfully captures contact events (bounces, friction-based deceleration, direction changes) that kinematic extrapolation cannot model. We introduce a centroid silhouette loss that isolates positional gradients from appearance noise, yielding 40% lower trajectory error than photometric supervision. We evaluate using cameras withheld from training that observe the object during its occlusion. Experiments on synthetic scenes show that PersistGS outperforms constant velocity extrapolation by +2.46dB PSNR and comes within 0.19dB of a ground-truth trajectory upper bound.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
