Near-Exhaustive Precomputation of Secondary Cloth Effects

How to read this

Category

Method: data-driven precomputation of secondary cloth motion

Contributions

• Massive offline exploration of the secondary cloth dynamics space over a large motion graph, using thousands of CPU-hours
• An incrementally constructed secondary cloth motion graph that samples the phase space to a low visual error tolerance
• Compression of tens of gigabytes of mesh data down to tens of megabytes for real-time interactive playback

Context

Challenges the curse-of-dimensionality argument against data-driven cloth and builds on reduced/learned clothing models such as de Aguiar et al.'s Stable Spaces for Real-time Clothing.Builds on: Stable Spaces for Real-time Clothing

Correctness

The approach assumes the character motion lives within a (large but bounded) motion graph and that the dynamical space can be sampled to a low visual-error tolerance; results are demonstrated as near-real-time playback rather than novel free-form motion, so generalization beyond the explored space is the limitation to keep in mind.

Clarity

Accessible framing; a first pass conveys the precompute-then-compress idea, a second pass pays off for the sampling and compression scheme.

How to read it

Focus on how the phase space is sampled and how the motion graph is compressed; a second pass is worth it to understand the error tolerance and storage tradeoff that make playback interactive.