Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PhyDeformer: High-Quality Non-Rigid Garment Registration with Physics-Awareness

Boyang Yu, Frederic Cordier, Hyewon Seo

TL;DR

PhyDeformer tackles high-fidelity non-rigid garment registration by coupling a coarse grading stage with a physics-aware, Jacobian-based refinement. The method first performs linear grading to align size and proportions, then optimizes per-triangle Jacobians $\mathbf{J}_i$ and solves a Poisson problem to produce a deformation $\phi$ that closely matches the target while respecting physical constraints. The objective combines reconstruction losses with membrane strain $\mathcal{L}_{s}$, bending $\mathcal{L}_{b}$, normal alignment $\mathcal{L}_{n}$, contour terms, and body-collision $\mathcal{L}_{c}$, yielding physically plausible deformations. Experiments on synthetic and real garments show superior geometric accuracy and efficiency compared to state-of-the-art methods, with demonstrated utility for inverse garment simulation. The approach offers a lightweight alternative to full physics-based simulation while delivering high-quality, wrinkle-level detail suitable for virtual try-on and related applications, though robustness to high-noise targets remains an area for future work.

Abstract

We present PhyDeformer, a new deformation method for high-quality garment mesh registration. It operates in two phases: In the first phase, a garment grading is performed to achieve a coarse 3D alignment between the mesh template and the target mesh, accounting for proportional scaling and fit (e.g. length, size). Then, the graded mesh is refined to align with the fine-grained details of the 3D target through an optimization coupled with the Jacobian-based deformation framework. Both quantitative and qualitative evaluations on synthetic and real garments highlight the effectiveness of our method.

PhyDeformer: High-Quality Non-Rigid Garment Registration with Physics-Awareness

TL;DR

PhyDeformer tackles high-fidelity non-rigid garment registration by coupling a coarse grading stage with a physics-aware, Jacobian-based refinement. The method first performs linear grading to align size and proportions, then optimizes per-triangle Jacobians and solves a Poisson problem to produce a deformation that closely matches the target while respecting physical constraints. The objective combines reconstruction losses with membrane strain , bending , normal alignment , contour terms, and body-collision , yielding physically plausible deformations. Experiments on synthetic and real garments show superior geometric accuracy and efficiency compared to state-of-the-art methods, with demonstrated utility for inverse garment simulation. The approach offers a lightweight alternative to full physics-based simulation while delivering high-quality, wrinkle-level detail suitable for virtual try-on and related applications, though robustness to high-noise targets remains an area for future work.

Abstract

We present PhyDeformer, a new deformation method for high-quality garment mesh registration. It operates in two phases: In the first phase, a garment grading is performed to achieve a coarse 3D alignment between the mesh template and the target mesh, accounting for proportional scaling and fit (e.g. length, size). Then, the graded mesh is refined to align with the fine-grained details of the 3D target through an optimization coupled with the Jacobian-based deformation framework. Both quantitative and qualitative evaluations on synthetic and real garments highlight the effectiveness of our method.
Paper Structure (13 sections, 7 equations, 8 figures, 2 tables)

This paper contains 13 sections, 7 equations, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Method
  4. Coarse grading
  5. Non-rigid geometry refinement
  6. Results and Experiments
  7. Implementation details
  8. Evaluation on synthetic garments
  9. Evaluation on 3D scan data
  10. Use case: Registration for inverse garment simulation
  11. Ablation study
  12. Robustness to noisy targets
  13. Conclusion and limitation

Figures (8)

  • Figure 1: Given an input 3D base mesh and a target garment, we first perform linear grading on the base mesh to achieve an initial alignment. It is further refined by optimizing per-triangle Jacobians.
  • Figure 2: Qualitative comparison of 3D garment reconstruction using our method versus other approaches yu2024inversede2023drapenetli2024isp.
  • Figure 3: Qualitative results on GarmCap dataset. Leftmost column: posed templates, second column: coarse fittings, third column: refined fittings by PhyDeformer, Rightmost column: target shape. Best viewed zoomed-in.
  • Figure 4: We evaluate the Chamfer Distance error for hybrid and IGPM yu2024inverse. On the right, we also illustrate the qualitative results for comparison: (a) Target; (b) New target by PhyDeformer; (c) IGBM with Hybrid scheme, and (d) IGPM.
  • Figure 5: Results of optimizing vertex displacements instead of Jacobians. From left to right: (a) source mesh, (b) deformed mesh with naive vertex-displacements optimization, (c) deformed mesh with clipped gradients to avoid holes caused by "NaNs", (d) target mesh.
  • ...and 3 more figures