Affinification: A Fine Approximation of Deformations

How to read this

Category

Academic Eurographics paper on physics based simulation, a model reduction method for elastic solids

Contributions

• Introduces affinification, a technique that automatically partitions an elastic solid into affine, cheap and low degree of freedom, and fully elastic, expensive and detailed, regions during simulation based on measured deformation rates
• Proposes a new clustering method to detect affinely deforming regions on the fly, plus heuristics for seeding and pattern generation of the coarsened regions
• Demonstrates up to 17 times speedup over full elastic simulation while keeping visual fidelity close to ground truth, with performance improving as mesh resolution increases
• Targets CPU only pipelines specifically, useful when GPU access is limited, for example mobile platforms or when the GPU is already busy with rendering

Context

The paper sits in the runtime adaptive model reduction line of physics based animation, the alternative to h and p refinement approaches that only add detail rather than coarsen it. It has no builds_on entry in the archive, but its own introduction positions it against prior derefinement and affine body work, arguing existing methods either fix the boundary or pay too much for remeshing.

Correctness

The method is compared against ground truth full elastic simulation, with reported speedups, mean 3.85 times on a toothbrush example, up to 17 times overall, and visual fidelity claims backed by side by side renders. This is a full peer reviewed Eurographics paper with quantitative benchmarks, not a talk.

Clarity

Dense, this is a from scratch academic method paper with its own clustering algorithm and physical parameter analysis, expect simulation math. Most useful for an engineer building or tuning a cloth or soft body solver rather than a generalist rigger.

How to read it

First pass, read the abstract and the toothbrush figure to see what affinification buys you and where. Second pass, read the introduction's positioning against refinement and derefinement approaches, and skim the clustering method section for the core idea. Third pass, worth the full read only if you are implementing a similar reduced order solver, work through the seeding and coarsening heuristics and the performance analysis by resolution.