libigl 🤝 TinyAD

A small example project mixing libigl and TinyAD. Launches the libigl viewer while in a separate thread optimizing an armadillo mesh's parametrization.

Compile

Compile this project using the standard cmake routine:

mkdir build
cd build
cmake ..
make

This should find and build the dependencies and create a example binary.

Run

From within the build directory just issue:

./example

A glfw app should launch displaying an animating Armadillo parametrization.

Derived from parametrization_libigl.cc

Ordering

TinyAD operates most conveniently on nodal vector values (e.g., vertex positions in V). If V contains nodal vectors per row, then regardless of whether V is stored as column-major (e.g., Eigen::MatrixXd) or row-major (e.g., Eigen::Matrix<double,Eigen::Dynamic,Eigen::Dynamic,Eigen::RowMajor>) the internal order of tiny-ad will correspond to non-standard row-major vectorization.

That is, if

V = [
  x₀ y₀ z₀
  x₁ y₁ z₁
  …
  xₙ yₙ zₙ
  ]

Then TinyAD will vectorize this into

x = [
  x₀
  y₀
  z₀
  x₁
  y₁
  z₁
  …
  xₙ
  yₙ
  zₙ
  ]

And use a corresponding ordering for gradients and Hessians.

This is ignorable if you're using the provided func.x_from_data and func.x_to_data. However, if you're mixing in your own gradients, Hessians, constraint projections, subspace bases, then take care!