Research map / 3D vision / Face and shape
Reconstruction of Manipulated Garment with Guided Deformation Prior
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.
Reconstructs 3D meshes of manipulated (folded or flattened) garments from incomplete point clouds by mapping points to UV space, fitting an ISP model, and completing UV maps with a guided diffusion deformation prior.
Problem
Recovering 3D garment shape from incomplete point clouds when garments are folded or crumpled rather than worn on a body, where self-occlusion and large non-rigid deformations make reconstruction hard.
Approach
Extends the Implicit Sewing Patterns (ISP) garment model with a DDPM-based deformation prior over UV maps and a UV mapper that projects point clouds into panel UV space; the diffusion model is guided by observed points and ISP panel masks to fill missing UV regions without requiring a garment template or body model.
Key results
- On VR-Folding Shirt folding, A3 of 84.7% vs 29.8% for GarmentTracking (GT init) and D_cf of 0.77 cm vs 1.16 cm.
- Outperforms GarmentTracking across Pants, Top, Shirt, Skirt for folding and flattening without requiring ground-truth initialization.
- Qualitatively reconstructs real-world pants and sweater point clouds captured with nerfstudio despite training on simulated data.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use the proposed reconstruction method instead of GarmentTracking with ground-truth initialization | VR-Folding garment reconstruction · Shirt, Pants, Top, and Skirt — Folding | A_3: 29.8% -> 84.7% (Shirt); 47.3% -> 75.9% (Pants); 37.9% -> 71.2% (Top); 23.5% -> 32.5% (Skirt) | Helped |
| Use the proposed reconstruction method instead of GarmentTracking with ground-truth initialization | VR-Folding garment reconstruction · Shirt, Pants, Top, and Skirt — Folding | D_cf: 1.16 -> 0.77 (Shirt); 1.07 -> 0.70 (Pants); 0.99 -> 0.74 (Top); 1.33 -> 1.04 (Skirt) | Helped |
| Use the proposed reconstruction method instead of GarmentTracking with ground-truth initialization | VR-Folding garment reconstruction · Shirt, Pants, Top, and Skirt — Flattening | A_5: 30.7% -> 78.8% (Shirt); 31.3% -> 60.8% (Pants); 54.6% -> 70.2% (Top); 5.4% -> 5.1% (Skirt) | Mixed |
| Use the proposed reconstruction method instead of GarmentTracking with ground-truth initialization | VR-Folding garment reconstruction · Shirt, Pants, Top, and Skirt — Flattening | D_cf: 1.78 -> 1.08 (Shirt); 1.64 -> 1.16 (Pants); 1.15 -> 1.08 (Top); 2.02 -> 1.75 (Skirt) | Helped |
Surprising findings
- GarmentTracking's accuracy drops substantially when initialized with network-predicted meshes (GN) instead of ground truth.
Limitations
- Uses a single frame for panel fitting and does not yet accumulate temporal point-cloud information
- Skirt metrics are ill-defined due to rotational symmetry
- Trained per garment category on simulated data
Experimental setup
Datasets: VR-Folding CLOTH3D
Benchmarks: VR-Folding
Metrics: A3 A5 A10 D_cr (correspondence distance) D_cf (Chamfer distance)
Baselines: GarmentNets GarmentTracking
Builds on: Implicit Sewing Patterns DDPM GarmentTracking GarmentNets
Closest papers by meaning
- Garment Recovery with Shape and Deformation Priors (CVPR 2024 · similarity 0.92)
- DressWild: Feed-Forward Pose-Agnostic Garment Sewing Pattern Generation from In-the-Wild Images (NeurIPS 2026 · similarity 0.89)
- VS: Reconstructing Clothed 3D Human from Single Image via Vertex Shift (CVPR 2024 · similarity 0.88)
- GaPT-DAR: Category-level Garments Pose Tracking via Integrated 2D Deformation and 3D Reconstruction (CVPR 2025 · similarity 0.87)
- DeClotH: Decomposable 3D Cloth and Human Body Reconstruction from a Single Image (CVPR 2025 · similarity 0.87)
- Learning 3D Garment Animation from Trajectories of A Piece of Cloth (NeurIPS 2024 · 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.