Restoring high-frequency wrinkles without full-scale simulation

Develop a method to restore realistic high-frequency wrinkles on 3D garment meshes generated by the SwiftTailor framework—specifically, meshes produced from the Garment Geometry Image by the GarmentSewer module—while avoiding reliance on full-scale physics-based sewing or draping simulation.

Background

SwiftTailor is a two-stage pipeline that predicts sewing patterns using the PatternMaker multimodal model and constructs 3D garments via GarmentSewer using the proposed Garment Geometry Image representation. This approach achieves fast, simulation-free garment construction with strong fidelity and stability.

However, the authors note that the reconstructed meshes lack high-frequency wrinkle detail because GarmentSewer tends to smooth fine geometric variations to ensure stable reconstruction. They explicitly state that recovering realistic high-frequency wrinkles remains an open problem and suggest that lightweight physics-based refinement or learning-based methods might address this without resorting to full-scale simulation.