kidchi/rigid-origami-crane-starter

lezhelun Rigid Origami State Set — paper crane, starter release

Produced by lezhelun / 乐折轮, a rigid-origami authoring engine.

A small, carefully measured set of rigid origami configurations of a single sheet (a traditional paper-crane crease pattern), released so that anyone can build and benchmark geometry-aware pipelines: 3D reconstruction, facet segmentation, crease-angle regression, shape tracking, and differentiable/kinematic solvers.

Every state in this set is a kinematically valid configuration of one and the same sheet: the paper is treated as an inextensible, unstretchable surface, the facets are rigid, and the creases are ideal hinges. Nothing here is hand-drawn or eyeballed — each state comes out of an in-house rigid-origami solver that was driven along a real folding sequence, and each state was checked for consistency before being exported.

What is inside

data/fold/S0001.fold.json      one file per state, standard FOLD format, 3D vertices
data/angles/S0001.csv          one row per crease: fold type, dihedral, angle away from flat
data/images/silhouette/S0001/  pure white-on-black shape, no lines, no texture
data/images/id/S0001/          one flat colour per facet — segmentation ground truth
metadata/states.csv            index of all states
tools/load_example.py          minimal loader

• 24 states, 6 camera views each, 2 image modes each → 288 images at 640×640.
• Views are placed on a ring around the sheet at two elevation bands (see the per-state azimuth/elevation recorded in the view names).
• The crease-angle table reports, for every crease, the dihedral angle between the two adjacent facets and the angle folded away from flat (0° = still flat, 180° = folded fully back). Creases that lie on the paper edge are marked paper boundary (no dihedral exists).

Coordinate and unit conventions

• Units are millimetres. The sheet is 155 mm × 155 mm.
• vertices_coords in the FOLD files are 3D and expressed in the same world frame for every state in a series; the camera record for each image is derivable from the view name plus the framing convention (a 22° half field of view, sheet inscribed in frame).
• Facet colours in the id images are assigned per material region, so the same piece of paper keeps the same colour in every state of this release. They are label values, not appearance.

What it is for

• pretraining a segmentation or detection backbone on clean, exactly-labelled geometry;
• validating a crease-angle estimator or a hinge-constraint solver against exact values;
• sanity-checking a shape tracker on single-colour silhouettes;
• teaching/benchmarking rigid-origami kinematics code that reads FOLD.

What it is not

• Not a photo dataset. The images are synthetic, untextured and noise-free. Do not use them as the only training source for a model that must run on camera footage.
• Not a full fold-sequence dataset. States are presented as an unordered set. This release deliberately does not describe how the states were ordered or produced.
• Not physically simulated. No elasticity, thickness, friction, gravity or collision response is modelled. Creases are ideal hinges and facets are perfectly rigid.
• No hands, no occlusion, no lighting variation, no background.

Provenance

Generated from a single traditional paper-crane crease pattern with lezhelun (乐折轮), a rigid-origami authoring engine. Every state in this set was produced by driving that engine along a real folding sequence and then checking the resulting configuration for consistency before export.

What is released here is the output geometry, not the means of producing it. The engine itself, the fold-choreography data, the material/topology source files and the authoring toolchain are not part of this release and are not required in order to use it.

Consistency checks performed before export:

• facets are rigid to machine precision along the sequence used to produce each state;
• shared crease endpoints of adjacent facets agree to better than 1e-9 mm;
• every crease in every state resolves to one of: a measured dihedral angle, or an explicit "on the paper boundary, no dihedral exists" marker. There are no missing values.

Known limitations

• The facet count is small (tens, not thousands). Flow-style dense deformations are out of scope.
• A crease reported as 0° may be a genuinely flat crease, or a crease that lies inside a region the solver keeps as one rigid plane. The status column distinguishes the cases.
• Only one crease pattern (paper crane) and one sheet size are included.

Licence

Released under Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0). You may use, share and adapt it for non-commercial purposes with attribution. Commercial use requires a separate licence — please get in touch.

See LICENSE and NOTICE.md.

Citation

If this set is useful in your work, please cite it as:

@misc{lezhelun_rigid_origami_state_set_crane_starter,
  title        = {lezhelun Rigid Origami State Set - paper crane, starter release},
  author       = {{lezhelun}},
  year         = {2026},
  note         = {Synthetic multi-view rigid-origami configurations with exact crease
                  angles, in FOLD format. Generated with the lezhelun authoring engine.},
  howpublished = {Hugging Face Datasets}
}

When you use the set, the attribution line we ask for is:

Dataset generated with lezhelun (乐折轮).

Feedback

Issues, requests for a larger split, or commercial licensing enquiries are welcome through the repository's discussion tab, or by e-mail:

kidchi.xyz@qq.com

Commercial licensing, datasets generated from your own crease pattern, and collaborations are handled through the same address.