FreeGave: 3D Physics Learning from Dynamic Videos by Gaussian Velocity
About
In this paper, we aim to model 3D scene geometry, appearance, and the underlying physics purely from multi-view videos. By applying various governing PDEs as PINN losses or incorporating physics simulation into neural networks, existing works often fail to learn complex physical motions at boundaries or require object priors such as masks or types. In this paper, we propose FreeGave to learn the physics of complex dynamic 3D scenes without needing any object priors. The key to our approach is to introduce a physics code followed by a carefully designed divergence-free module for estimating a per-Gaussian velocity field, without relying on the inefficient PINN losses. Extensive experiments on three public datasets and a newly collected challenging real-world dataset demonstrate the superior performance of our method for future frame extrapolation and motion segmentation. Most notably, our investigation into the learned physics codes reveals that they truly learn meaningful 3D physical motion patterns in the absence of any human labels in training.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Future frame extrapolation | Dynamic Indoor Scene Dataset | PSNR35.019 | 24 | |
| Future frame extrapolation | Dynamic Object Dataset | PSNR31.987 | 22 | |
| Novel view interpolation | Dynamic Indoor Scene Dataset | PSNR32.287 | 22 | |
| Novel view interpolation | Dynamic Object Dataset | PSNR39.393 | 20 | |
| Future frame extrapolation | NVIDIA Dynamic Scene Truck | PSNR29.954 | 12 | |
| Novel view interpolation | NVIDIA Dynamic Scene Truck | PSNR28.584 | 12 | |
| Future frame extrapolation | NVIDIA Dynamic Scene Skating | PSNR28.391 | 12 | |
| Novel view interpolation | NVIDIA Dynamic Scene Skating | PSNR26.589 | 12 | |
| Future frame extrapolation | Dynamic Multipart (test) | PSNR34.667 | 9 | |
| Unsupervised Object Segmentation | synthetic indoor scene dataset | AP99.75 | 7 |