Research map / 3D vision / 3D generation
Physically Compatible 3D Object Modeling from a Single Image
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes physical compatibility optimization that treats rest-shape geometry as the variable under a static-equilibrium FEM hard constraint, so single-image 3D reconstructions become stable and match the input image under gravity.
Problem
Single-image 3D reconstruction methods ignore physics, so reconstructed objects topple, sag, or deform under gravity and diverge from the depicted static geometry.
Approach
Decomposes an object's visual geometry into mechanical properties, external forces, and rest-shape geometry, and optimizes the rest shape (parameterized by a plastic deformation field) subject to static equilibrium enforced as a hard constraint via implicit differentiation, so the physical object matches the image or stays upright.
Key results
- TetSphere with the framework reaches 92.2% standability versus 32.8% baseline and lowers mean stress from 0.22 to 0.19 kPa.
- Wonder3D standability rises from 6.9% to 72.4% with the framework, with mean stress dropping from 10.68 to 0.45 kPa.
- Across all five baselines, the framework improves image loss, mean stress and standability on 100 Objaverse shapes.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Apply physical compatibility optimization | Wonder3D · Objaverse evaluation set | Mean Stress (kPa): 10.68 ± 17.47 -> 0.45 ± 0.96 | Helped |
| Apply physical compatibility optimization | LGM · Objaverse evaluation set | Mean Stress (kPa): 1.14 ± 2.03 -> 1.01 ± 1.34 | Helped |
| Apply physical compatibility optimization | MeshLRM · Objaverse evaluation set | Mean Stress (kPa): 0.54 ± 1.41 -> 0.38 ± 1.05 | Helped |
| Apply physical compatibility optimization | TripoSR · Objaverse evaluation set | Mean Stress (kPa): 0.29 ± 1.28 -> 0.22 ± 0.94 | Helped |
| Apply physical compatibility optimization | TetSphere · Objaverse evaluation set | Mean Stress (kPa): 0.22 ± 0.51 -> 0.19 ± 0.78 | Helped |
| Apply physical compatibility optimization | Wonder3D · Objaverse evaluation set | Standable (%): 6.9 -> 72.4 | Helped |
| Apply physical compatibility optimization | LGM · Objaverse evaluation set | Standable (%): 20.3 -> 85.5 | Helped |
| Apply physical compatibility optimization | MeshLRM · Objaverse evaluation set | Standable (%): 29.6 -> 74.5 | Helped |
| Apply physical compatibility optimization | TripoSR · Objaverse evaluation set | Standable (%): 24.2 -> 80.6 | Helped |
| Apply physical compatibility optimization | TetSphere · Objaverse evaluation set | Standable (%): 32.8 -> 92.2 | Helped |
| Apply physical compatibility optimization | Wonder3D · Objaverse evaluation set | Img. Loss: 0.073 ± 0.063 -> 0.069 ± 0.048 | Helped |
| Apply physical compatibility optimization | LGM · Objaverse evaluation set | Img. Loss: 0.121 ± 0.091 -> 0.116 ± 0.065 | Helped |
Limitations
- Requires user-specified material properties and external forces rather than inferring them from the image.
- Relies on tetrahedral meshes obtained by converting baseline outputs, with no differentiable converter.
- Only handles static equilibrium, not dynamic motion.
Experimental setup
Datasets: Objaverse
Benchmarks: Objaverse-derived evaluation set of 100 shapes
Metrics: number of connected components mean stress standability image loss fracture rate
Baselines: Wonder3D LGM MeshLRM TripoSR TetSphere
Builds on: Neo-Hookean material model TetSphere TetWild Adam
Closest papers by meaning
- Ghost on the Shell: An Expressive Representation of General 3D Shapes (ICLR 2024 · similarity 0.87)
- OmniPhysGS: 3D Constitutive Gaussians for General Physics-Based Dynamics Generation (ICLR 2025 · similarity 0.86)
- Vid2Sim: Generalizable, Video-based Reconstruction of Appearance, Geometry and Physics for Mesh-free Simulation (CVPR 2025 · similarity 0.86)
- Simulation-Ready Compositional 3D Scene Reconstruction from a Single Image (NeurIPS 2026 · similarity 0.86)
- PhysX-3D: Physical-Grounded 3D Asset Generation (NeurIPS 2025 · similarity 0.86)
- MonoPhysics: Estimating Geometry, Appearance, and Physical Parameters from Monocular Videos (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.