Multiscaling and landmarks

The skshapes.Multiscale class preserves landmarks across scales.

import pyvista as pv

import skshapes as sks

To demonstrate this, we first load a mesh with a collection of key points.

mesh = sks.PolyData(pv.examples.download_louis_louvre().clean())

landmarks = [151807, 21294, 23344, 25789, 131262, 33852, 171465, 191680]
landmarks += [172653, 130895, 9743, 19185, 143397, 200885]

mesh.landmark_indices = landmarks

Then, we compute a multiscale representation of our mesh using 10%, 1% and 0.1% of the original point count.

ratios = [1, 0.1, 0.01, 0.001]
multiscale = sks.Multiscale(shape=mesh, ratios=ratios)

The landmark_points and landmark_indices attributes of the shape are transported consistently between the different scales.

pl = pv.Plotter()

pl.open_gif("animation.gif", fps=1)
for ratio in ratios:
    mesh_i = multiscale.at(ratio=ratio)
    print(f"with {mesh_i.n_points:,d} points, landmarks = ")
    print([int(i) for i in mesh_i.landmark_indices])

    pl.clear_actors()
    sks.doc.display(plotter=pl, shape=mesh_i)
    sks.doc.display(plotter=pl, shape=mesh_i.landmark_points, color="red")
    pl.camera_position = "xz"
    pl.camera.zoom(1.4)
    pl.write_frame()

pl.show()

with 210,873 points, landmarks =
[151807, 21294, 23344, 25789, 131262, 33852, 171465, 191680, 172653, 130895, 9743, 19185, 143397, 200885]
with 21,067 points, landmarks =
[14747, 1615, 2465, 10164, 16549, 3281, 16824, 18981, 16971, 5351, 943, 1778, 13775, 16290]
with 2,087 points, landmarks =
[144, 181, 205, 203, 1630, 336, 1659, 1690, 1090, 1202, 73, 157, 1917, 1566]
with 188 points, landmarks =
[19, 84, 20, 17, 28, 27, 147, 33, 32, 184, 12, 126, 169, 137]