Skip to content

Examples

Flood PH of noisy torus points

In the following example, we sample n_pts=1000000 points from a noisy torus in 3D, then construct the Flood complex on top of n_lms=1000 landmarks (see Illustration ) and compute persistent homology (up to dimension 3) using gudhi based on the constructed filtered Flood simplicial complex.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
from flooder import (
    generate_noisy_torus_points_3d, 
    flood_complex, 
    generate_landmarks)

DEVICE = "cuda"
n_pts = 1_000_000  # Number of points to sample from torus
n_lms = 1_000      # Number of landmarks for Flood complex

pts = generate_noisy_torus_points_3d(n_pts).to(DEVICE)
lms = generate_landmarks(pts, n_lms)

stree = flood_complex(pts, lms, return_simplex_tree=True)
stree.compute_persistence()
ph_diags = [stree.persistence_intervals_in_dimension(i) for i in range(3)]

Importantly, one can either call flood_complex with the already pre-selected (here via FPS) landmarks, or one can just specify the number of desired landmarks, e.g., via

1
stree = flood_complex(pts, n_lms, return_simplex_tree=True)

in which case FPS is called internally.

Flooder CLI

If you installed flooder via pip install flooder, you not only have the API available but also a command-line interface (CLI), which you can execute as follows: for demonstration, we will download a PLY file of the Lucy angel (>14M points, available from the Stanford 3D Scanning Repository) 

1
2
wget https://graphics.stanford.edu/data/3Dscanrep/lucy.tar.gz
tar xvfz lucy.tar.gz

and first convert it (using the Open3D library; install via pip install open3d) to a (N,3) numpy array:

1
2
3
4
5
import numpy as np
import open3d as o3d
X_ply = o3d.io.read_point_cloud("lucy.ply")
X_np = np.asarray(X_ply.points, dtype=np.float32)
np.save("lucy.npy", X_np)

Finally, we execute the Flooder CLI, using 5k landmarks, 30 points per edge and a batch size of 64:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
flooder \
  --input-file lucy.npy \
  --output-file lucy-diagrams.pkl \
  --num-landmarks 5000 \
  --batch-size 64 \
  --max-dimension 3 \
  --points-per-edge 30 \
   --device cuda:0 \
  --cuda-events \
  --stats-json lucy.json