drone12/creasepattern

Origami Crease Pattern Dataset

The accompanying h5df file comprises the dataset of 8 million crease patterns used for the training of a VAE with normalizing flows as described in the paper: "Inverse Design of Origami for Trajectory Following", Journal of Mechanisms and Robotics, 2025.

Authors

• Nicolas Hochuli (a), nhochuli@ethz.ch
• Tino Stankovic (a,1), tinos@ethz.ch

(a) Engineering Design and Computing Laboratory, Department of Mechanical and Process Engineering, ETH Zurich, Tannenstrasse 3, 8092 Zurich, Switzerland
(1) To whom correspondence should be addressed

Usage

The h5df file dataset.hdf5 can be opened in Python using the pip-installable h5py package. The h5df format is a widely used format for storing large data in single files. Basic usage is described in docs.h5py.org

The file dataset.hdf5 holds the connectivity information (edge_lists, shape: [Ex2], E=Number of edges) and geometric vertex positions (vertex_coords, shape: [Nx2], N=Number of vertices) for all 8 million planar crease patterns in two separate groups.

File Structure

The hierarchical file structure in dataset.hdf5 is organized as described here. For efficiency reasons, the two groups edge_lists and vertex_coords are split into 400 subgroups each, where each subgroup holds 20000 datasets (i.e. samples / crease patterns). The tree structure is defined as:

dataset.hdf5        
├── metadata (attributes)
├── edge_lists/     ← Connectivity information (edge lists) of all 8 million crease patterns, in 400 subgroups.
│   ├── 0/          ← In a nested group, holds 20000 samples (as so-called datasets)
│   ├── 1/          ← In a nested group, holds 20000 samples (as so-called datasets)
│   ├── ...
│   └── 399/
└── vertex_coords/  ← Vertex position information (2D) of all 8 million crease patterns.
    ├── 0/
    ├── 1/
    ├── ...
    └── 399/

Example

To load the dataset into memory using Python, printing the metadata, retrieving the geometry of a single crease pattern and generating a visualization, run this code:

import h5py
import matplotlib.pyplot as plt

h5 = h5py.File("dataset.hdf5", "r")

# Example: Get creasepattern 18340 from subgroup 20
edge_list = h5["edge_lists/20/18340"][:,:]  # Numpy array (Ex2)
vertex_coords = h5["vertex_coords/20/18340"][:,:]  # Numpy array (Nx2)

# Print all metadata attributes 
print(h5["metadata"].attrs["Title"])
print(h5["metadata"].attrs["Journal"])
print(h5["metadata"].attrs["Authors"])
print(h5["metadata"].attrs["Year"])  
print(h5["metadata"].attrs["Size"]) 
print(h5["metadata"].attrs["Version"])

### Create plot
plt.figure(figsize=(8, 6))
for edge in edge_list:
    p1 = vertex_coords[edge[0]]
    p2 = vertex_coords[edge[1]]
    plt.plot([p1[0], p2[0]], [p1[1], p2[1]], "k-", lw=1)

plt.scatter(vertex_coords[:, 0], vertex_coords[:, 1], color="blue", zorder=3)
for idx, (x, y) in enumerate(vertex_coords):
    plt.text(x, y, str(idx), fontsize=8, ha="right", va="bottom")

plt.axis("equal")
plt.title("Crease Pattern Visualization")
plt.grid(True)
plt.tight_layout()
plt.show()