MonoPhysics: Estimating Geometry, Appearance, and Physical Parameters from Monocular Videos

Daniel Rho1, Jun Myeong Choi1, Matthew Thornton1, Biswadip Dey2, and Roni Sengupta1

1University of North Carolina at Chapel Hill  ·  2Meta

arXiv Dataset Code
Overview of MonoPhysics pipeline

MonoPhysics recovers geometry, appearance, and material parameters of deformable objects from a single monocular video.

To make this work, we propose three new techniques:

  1. Scene Re-parameterization.
    A single learnable scalar resolves monocular scale ambiguity.
  2. Physics-aware Geometric Refinement.
    Simulation feedback guides where Gaussians move and how much space they occupy.
  3. Differentiable Position Map.
    Pixel-location losses, such as silhouettes, flow back to particle positions.

Results

Seen Unseen (future prediction)

Vid2Sim

Our Dataset

Elastic

Plasticine

BibTeX

@article{rho2026monophysics,
  title={MonoPhysics: Estimating Geometry, Appearance, and Physical Parameters from Monocular Videos},
  author={Rho, Daniel and Choi, Jun Myeong and Thornton, Matthew and Dey, Biswadip and Sengupta, Roni},
  journal={arXiv preprint arXiv:2605.30320},
  year={2026}
}