Learning-Based Animation of Clothing for Virtual Try-On

How to read this

Category

Method: learning-based clothing animation for virtual try-on

Contributions

• Trains a learned cloth model from a database of physically based dressed-character simulations across many body shapes and animations
• Separates global garment fit (from body shape) from local wrinkles (from pose dynamics and body shape)
• Uses a recurrent network to regress nonlinear wrinkles and produces try-on animations in a few milliseconds, avoiding prior blending artifacts

Context

Combines a parametric body model with learned cloth deformation, building on Loper et al.'s 'SMPL' (2015) and the real-time clothing-space idea of de Aguiar et al.'s 'Stable Spaces for Real-time Clothing' (2010).Builds on: SMPL: A Skinned Multi-Person Linear Model · Stable Spaces for Real-time Clothing

Correctness

The fit-versus-wrinkle decomposition and RNN dynamics give plausible, fast results, but quality is bounded by the precomputed simulation database covering the relevant body shapes and motions, and it targets a fixed garment per trained model.

Clarity

Clear structure and well-motivated; a first pass conveys the fit/wrinkle split, while the RNN formulation and training setup reward a second pass.

How to read it

First pass for the global-fit-plus-local-wrinkle decomposition and the millisecond runtime claim; second pass on the RNN and the simulation-database construction if you plan to train or extend it.