Research map / 3D vision / Gaussian splatting
REArtGS: Reconstructing and Generating Articulated Objects via 3D Gaussian Splatting with Geometric and Motion Constraints
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 REArtGS, a 3D Gaussian Splatting framework that adds an unbiased SDF guidance for opacity-field surface reconstruction and kinematic-constrained deformable fields to generate meshes of unseen articulation states from two RGB states.
Problem
Reconstructing high-fidelity textured surface meshes of articulated objects and generating meshes at unseen joint states, given multi-view RGB images from only two arbitrary states, without intermediate-state supervision.
Approach
Couples an SDF field with Gaussian opacity via a bell-shaped SDF-to-opacity function and an unbiased regularizer that pins the SDF to zero at each Gaussian's maximum-contribution depth; deformation fields for Gaussians are parameterized by revolute (Rodrigues rotation) or prismatic motion with an unsupervised part segmentation.
Key results
- On PartNet-Mobility reconstruction, REArtGS reports mean CD(ws) 3.79, CD(rs) 1.236, F1 0.294, EMD 0.695, outperforming the baselines in the mean.
- On PartNet-Mobility generation at unseen states, REArtGS reports mean CD(ws) 4.10 and CD(rs) 1.063, best among compared methods (ArtGS mean CD(ws) 4.92).
- On real-world AKB-48 reconstruction, REArtGS mean CD(ws) is 2.86 versus 6.41 for PARIS and 20.61 for ArtGS.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add SDF guidance (ablation) | REArtGS · PartNet-Mobility reconstruction | CD (ws): 5.96 -> 4.38 | Helped |
| add unbiased SDF regularization to SDF guidance (ablation) | REArtGS · PartNet-Mobility reconstruction | CD (ws): 4.38 -> 3.79 | Helped |
| add motion constraints (ablation) | REArtGS · PartNet-Mobility generation | CD (ws): 18.65 -> 5.41 | Helped |
Surprising findings
- ArtGS and PARIS yield very large CD(ws) errors on some AKB-48 categories (e.g., ArtGS 51.04 on Stapler), indicating instability on real data.
Limitations
- Requires known camera poses as prior input.
- Struggles with objects containing transparent materials.
- Assumes only rotation and translation motions and two observed states.
Experimental setup
Datasets: PartNet-Mobility AKB-48
Benchmarks: PartNet-Mobility AKB-48
Metrics: Chamfer Distance (CD) F1-score Earth Mover's Distance (EMD)
Baselines: A-SDF Ditto PARIS GOF ArtGS D-3DGS
Builds on: GOF Neus PARIS ArtGS 3D Gaussian Splatting
Closest papers by meaning
- ArticulatedGS: Self-supervised Digital Twin Modeling of Articulated Objects using 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- Building Interactable Replicas of Complex Articulated Objects via Gaussian Splatting (ICLR 2025 · similarity 0.90)
- REACTO: Reconstructing Articulated Objects from a Single Video (CVPR 2024 · similarity 0.89)
- RigGS: Rigging of 3D Gaussians for Modeling Articulated Objects in Videos (CVPR 2025 · similarity 0.89)
- FreeTimeGS: Free Gaussian Primitives at Anytime Anywhere for Dynamic Scene Reconstruction (CVPR 2025 · similarity 0.87)
- MotionGS: Exploring Explicit Motion Guidance for Deformable 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.87)
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.