Physics-Inspired Upsampling for Cloth Simulation in Games

How to read this

Category

Method: data-driven cloth upsampling for interactive/game use

Contributions

• Learns linear upsampling operators that add mid-scale detail to coarse cloth meshes at interactive performance
• Uses harmonic regularization to fit training data without overfitting and harmonic test functions as tracking constraints to align coarse and fine simulations
• Demonstrates generalization to unseen conditions such as different wind velocities and novel character motions

Context

Sits in the coarse-to-fine cloth detailing line and pairs with position-based / compliant-constraint simulators (XPBD, Macklin et al.) as the underlying low-resolution solver it enriches.Builds on: XPBD: Position-Based Simulation of Compliant Constrained Dynamics

Correctness

Linear operators keep the method fast and the harmonic constraints aim to curb overfitting, with generalization shown on wind and motion variation; being a learned upsampler it adds plausible mid-scale detail rather than physically exact fine-scale dynamics, so accuracy versus a full high-resolution sim is a caveat.

Clarity

Accessible framing for practitioners; a first pass conveys the upsampling idea and games target, a second pass is needed for the harmonic regularization and tracking-constraint math.

How to read it

First pass for whether learned upsampling fits your real-time pipeline; do a second pass on the harmonic test-function constraints if you need to train operators for your own cloth assets.