Lecture 11

Cloth simulation.

Thoughts

  • The first paper about making cloth sim really work had a surprising amount of non-physical impulses and corrections. The result was still pretty realistic (though not entirely there), so I wonder how far we can take these corrections before meeting to move to a fully yarn-based model to achieve better realism. I imagine this works well enough for production simulation, though, which prioritizes reliability.
  • Modelling cloth properties off of physical measurements is cool. These days I bet that would be a differentiable simulation/rendering problem.
  • Exhaustive search of cloth simulation clips was a cool idea, and resulted in real-time results good enough for games. The pre-computation time was huge, though.
  • Finally we get to the yarn-level simulation. Amazing!
  • And a differentiable rendering approach. Approximate matching using just a normal and texture-space displacement map was an interesting approach to de-coupling the geometry and material spaces.
Written on March 10, 2021