Search papers, labs, and topics across Lattice.
The paper introduces Motion Group-aware Gaussian Forecasting (MoGaF), a novel framework for long-term dynamic scene extrapolation using 4D Gaussian Splatting. MoGaF enforces physically consistent motion by introducing motion-aware Gaussian grouping and group-wise optimization, leading to spatially coherent dynamic representations. Experiments on synthetic and real-world datasets demonstrate that MoGaF outperforms existing baselines in rendering quality, motion plausibility, and long-term forecasting stability.
Achieve realistic and temporally stable scene evolution by enforcing physically consistent motion across both rigid and non-rigid regions using motion-aware Gaussian grouping.
Forecasting dynamic scenes remains a fundamental challenge in computer vision, as limited observations make it difficult to capture coherent object-level motion and long-term temporal evolution. We present Motion Group-aware Gaussian Forecasting (MoGaF), a framework for long-term scene extrapolation built upon the 4D Gaussian Splatting representation. MoGaF introduces motion-aware Gaussian grouping and group-wise optimization to enforce physically consistent motion across both rigid and non-rigid regions, yielding spatially coherent dynamic representations. Leveraging this structured space-time representation, a lightweight forecasting module predicts future motion, enabling realistic and temporally stable scene evolution. Experiments on synthetic and real-world datasets demonstrate that MoGaF consistently outperforms existing baselines in rendering quality, motion plausibility, and long-term forecasting stability. Our project page is available at https://slime0519.github.io/mogaf
