EUROGRAPHICS 2014 / B. Lévy and J.

Kautz Volume 33 (2014), Number 2

(Guest Editors)

Multi-style Paper Pop-up Designs from 3D Models

Conrado R. Ruiz Jr. Sang N. Le Jinze Yu Kok-Lim Low

Department of Computer Science, National University of Singapore

Figure 1: A multi-style paper pop-up constructed from a design layout that is automatically generated from an input 3D model.

Abstract
Paper pop-ups are interesting three-dimensional books that fascinate people of all ages. The design and con-
struction of these pop-up books however are done manually and require a lot of time and effort. This has led to
computer-assisted or automated tools for designing paper pop-ups. This paper proposes an approach for automat-
ically converting a 3D model into a multi-style paper pop-up. Previous automated approaches have only focused
on single-style pop-ups, where each is made of a single type of pop-up mechanisms. In our work, we combine
multiple styles in a pop-up, which is more representative of actual artist’s creations. Our method abstracts a 3D
model using suitable primitive shapes that both facilitate the formation of the considered pop-up mechanisms and
closely approximate the input model. Each shape is then abstracted using a set of 2D patches that combine to
form a valid pop-up. We define geometric conditions that ensure the validity of the combined pop-up structures. In
addition, our method also employs an image-based approach for producing the patches to preserve the textures,
finer details and important contours of the input model. Finally, our system produces a printable design layout and
decides an assembly order for the construction instructions. The feasibility of our results is verified by constructing
the actual paper pop-ups from the designs generated by our system.
Categories and Subject Descriptors (according to ACM CCS): I.3.5 [Computer Graphics]: Computational Geometry
and Object Modeling—Geometric algorithms, languages, and systems

1. Introduction Recently, there has been much interest in the physical fab-
rication of 3D models. Paper pop-ups are a practical candi-
Paper pop-ups or movable books are three dimensional date for this task since they do not require specialized hard-
books that contain paper pieces that pop out or move when ware and they can be folded flat for easy storage. Just as al-
the book is opened and fold completely flat when the book is gorithms in origami have found applications in protein fold-
closed. Although now popularly used for children’s books, it ing and deploying instruments in space, pop-up algorithms
was historically also employed for illustration purposes in a could potentially be used for other applications. Examples
wider range of topics like philosophy, astronomy, geometry include 3D micro-fabrication from 2D patterns and collapsi-
and medicine. Today’s pop-up books still continue to fasci- ble objects such as foldable furniture.
nate readers of all ages and cultures. Some of the notable
artists are Robert Crowther [Cro11], Robert Sabuda [CS03] Creating a pop-up however can be a tedious task even for
and David Carter [Car08] (Figure 2). an experienced designer. It usually entails a trial-and-error

c 2014 The Author(s)

Computer Graphics Forum c 2014 The Eurographics Association and John
Wiley & Sons Ltd. Published by John Wiley & Sons Ltd.
 C. R. Ruiz Jr., S. N. Le, J. Yu & K.-L. Low / Multi-style Paper Pop-up Designs from 3D Models

Figure 2: Sample pop-up books (left to right): Amazing Pop-up Trucks [Cro11], Alice’s Adventures in Wonderland [CS03] and
Yellow Squares [Car08].

approach to find a valid configuration of the pieces of paper, In the automated approaches of [LSH∗ 10] and [LJGH11],
so that they resemble a desired structure. A pop-up book may voxelization is used to approximate the input 3D model,
require weeks to design and up to a year to complete. With which leads to the possible loss of important features. In
the proliferation of 3D models on the web and the easy ac- our work, the shape abstraction allows us to fit a minimal
cessibility to 3D authoring software, we propose a fully au- number of 3D primitives to approximate the model, result-
tomated approach for converting 3D models into paper pop- ing in fewer patches. The final patches are produced using
ups. an image-based approach to preserve the textures, finer de-
tails and important contours of the input model. Finally, our
The only methods that are able to automatically generate system produces a printable design layout and decides an
pop-up designs are [LSH∗ 10], [LJGH11], [LNLRL13] and assembly order for the construction instructions.
[LLLN∗ 13]. In each of these works, each pop-up is made of
only a single type of pop-up mechanisms (i.e. single-style
pop-up), and a very specialized method is used to gener- 2. Related Work
ate the pop-up design. The v-style pop-ups addressed by
Papercraft. Since the invention of paper itself, paper craft-
[LJGH11] seem to be the most versatile in terms of geom-
ing has fascinated people and has been the subject of several
etry. However, the main focus of [LJGH11] was on the ge-
studies. Mitani and Suzuki [MS04] proposed an automatic
ometric study, and its automatic method can only generate
method for creating paper toys by simplifying 3D meshes
pop-up patches restricted to three perpendicular orientations.
into paper strips. Origami, the traditional Japanese art of pa-
As such, it is not able to demonstrate the full potential of the
per folding, is another popular type of papercraft that has
v-style mechanism.
been extensively studied in literature [Hul06,DO07,O’R11].
In actual pop-up books created by artists, numerous styles In particular, folding algorithms and conditions for foldabil-
are used to suitably represent different parts of the objects. ity are of much interest. Recently, Tachi [Tac10] proposed
Our objective is thus to combine multiple styles in a pop- an algorithm for designing origamic structures automatically
up, and use the most suitable mechanism for each part of the from a polyhedral surface. Although sharing one of the cen-
object. Combining multiple styles presents new challenges tral problems of pop-ups, which is foldability, the physical
in the validation of its stability and foldability. In this pa- constraints are significantly different for origami. As a result,
per, we provide new geometric conditions for the validity of their formulations cannot easily be used for paper pop-ups.
multi-style pop-ups. Mesh Simplification and Abstraction. Automatic generation
We consider four types of mechanisms, the step-fold, tent- of paper pop-up designs from 3D models can be consid-
fold, v-fold and box-fold. Of these mechanisms, the box-fold ered a form of geometry simplification. [LNLRL13] uses
has not yet been studied in any previous work. As such we generalized cylinders to approximate 3D models in order
formally include a description of the box-fold and outline to automatically generate lattice-style pop-ups or sliceforms
the conditions for its validity considering its foldability and designs. Our approach is inspired by the promising results
stability. in shape abstraction using 2D planes [DDSD03, MSM11]
and 3D primitives [YK12]. Our approximation aims to pre-
Our method abstracts the input 3D model using suitable serve the input contours, as they have been shown to cap-
primitive shapes that both facilitate the formation of the con- ture important shape information of the objects [MZL∗ 09].
sidered pop-up mechanisms and closely approximate the in- Nonetheless, the existing works in shape abstraction cannot
put model. Each shape is then abstracted using a set of 2D be readily applied to our domain, since an arbitrary set of 2D
patches that combine to form a valid pop-up. planes may not necessarily be a valid pop-up.

c 2014 The Author(s)
Computer Graphics Forum
c 2014 The Eurographics Association and John Wiley & Sons Ltd.
 C. R. Ruiz Jr., S. N. Le, J. Yu & K.-L. Low / Multi-style Paper Pop-up Designs from 3D Models

Figure 3: Pop-up mechanisms and the corresponding 3D primitives: (a) step-fold, (b) tent-fold, (c) box-fold and (d) v-fold. The
shaded faces are the principal faces.

In addition, artists consider the overall structure of an ob- 3. Pop-up Fundamentals

ject and its parts to determine which mechanisms to use. As
3.1. Terminology
such, our work is also related to semantic segmentation, like
the recent work of [vKXZ∗ 13] on the hierarchical analysis of A pop-up is created by folding and gluing together pieces of
shape structures. Other approaches are detailed in the work paper. It comprises a set of planar polygons called patches
of [MWZ∗ 13]. that are connected at straight line segments called hinges or
fold lines. A pop-up has to sit between two primary patches
Computational Pop-ups. A number of studies on paper pop- called the ground and backdrop, which are connected by a
ups have been presented. [LTS96, Gla02a, Gla02b] describe special hinge called the central fold. The interior angle be-
the use of simple geometry to model various pop-up styles. tween these two patches is called the fold angle.
[HE06, LTS96] present interactive design systems that sim-
ulate the opening and closing of paper pop-ups. Iizuka et al.
[IEM∗ 11] also presented an interactive system that supports
v-folds and parallel folds. Hoiem et al. [HEH05] developed
an application to create simple pop-ups from photographs
based on scene or image understanding [HEH08]. Abel et
al. [ADD∗ 12] proposed an algorithm for creating pop-ups
by subdividing an input polygon into a single-degree-of-
freedom linkage structure.

A few studies have focused particularly on Origamic Ar- Figure 4: Basic pop-up terminology.
chitectures (OA), a special type of paper pop-up that is made
of a single sheet of paper [MSU03,LSH∗ 10,LLLN∗ 13]. One We define a mechanism as the most basic geometric struc-
notable recent work is that of [LSH∗ 10], in which they pro- ture that, together with other structures of the same type,
vided a theoretical foundation and a voxel-based algorithm gives a pop-up a unique style. Pop-up artists use several
for automatic design of origamic architectures. types of mechanisms to make their works come to life. In
this paper, we focus on four mechanisms: step-fold, tent-
[LJGH11] later extended the notion of validity to a more fold, v-fold, and box-fold. Among these mechanisms, step-
general class called v-style pop-ups. They also developed an fold structure pops up at 90◦ fold angle, while the others pop
interactive tool to convert a 3D model to a v-style paper pop- up at 180◦ . Illustrations of these mechanisms and their 3D
up. Similar to [LSH∗ 10], they used voxelization to approx- primitive pairs, which are used in our algorithm, are shown
imate the input model, which may cause important features in Figure 3.
to be lost. In order to capture accurate contours, a very dense
voxel grid is required, which then makes the design infeasi- When creating a pop-up, we also require it to be valid,
ble to construct in practice because of the excessive amount which means it needs to be both foldable and stable. A pop-
of cuts and folds. In addition, their v-style pop-ups are lim- up is said to be foldable if the structure can fold completely
ited to only three patch orientations. flat when the ground and backdrop patches are fully closed.
Note that during the folding process, the rigidity and connec-
Note that most of the computer-aided design systems be- tivity of the patches need to be maintained at all times and
fore [LSH∗ 10, LJGH11] are limited to a few known mech- it should not introduce new fold lines. On the other hand,
anisms, and do not guarantee the validity of the designs. a pop-up is said to be stable if all its patches are stationary
Research on the geometric conditions for the validity of a when the ground and backdrop patches are held still at any
general paper pop-up is scarce, primarily because determin- fold angle. In other words, the closing and opening of a pop-
ing the foldability of a general pop-up is NP-hard, as shown up do not need any extra external force besides holding the
in [UT06]. two primary patches.

c 2014 The Author(s)

Computer Graphics Forum
c 2014 The Eurographics Association and John Wiley & Sons Ltd.
 C. R. Ruiz Jr., S. N. Le, J. Yu & K.-L. Low / Multi-style Paper Pop-up Designs from 3D Models

Figure 5: Pop-up mechanisms and patches: (a) step-fold, (b) tent-fold, (c) type-1 v-fold and (d) type-2 v-fold.

To facilitate our formulation, we assume that paper is The second type of v-fold (Figure 5d) is similar to the first
rigid and has no thickness or weight. These assumptions type, except that it forms only a triangular half of a box when
are similar to those used in the previous computational stud- B and G are fully opened at 180◦ , and it does not have the
ies of paper art [BH02, LSH∗ 10, LJGH11]. We also use the back patches BL and BR .
same formal definitions of foldability and stability as used
3.2.4. Box-fold
in [LSH∗ 10, LJGH11].
Unlike the other pop-up mechanisms, box-fold has yet to be
formulated in detail in previous work. Box-fold is the most
3.2. Pop-up Mechanisms complex of the four mechanisms considered. We choose to
In this section, we describe the patch arrangement and con- include it because of its ability to capture rectangular objects
straints in the considered pop-up mechanisms. aligned with the central fold line and its common usage in
many artworks, such as [Cro11].
3.2.1. Step-fold
Step-fold, which is referred to as mechanism D2 in
[LJGH11], comprises 4 patches, S1 , S2 , S3 , and S4 , folded
together such that S1 is parallel to S3 and S2 is parallel to S4
(Figure 5a). Patches S3 and S4 are bounded by their cut lines
and common hinge.

3.2.2. Tent-fold
Tent-fold, as described in [HS09], is more general than step- Figure 6: The box-fold mechanism.
fold in that it does not require its patches T1 , T2 , T3 and T4 to Consider a pop-up opened at 180◦ . A box-fold structure
form parallel pairs, although the fold lines need to be paral- comprises 11 patches labeled as G, B, L, R, FL , FR , BL , BR ,
lel. We construct tent-fold so that |T1 | = |T2 |, |T3 | = |T4 | and TG , TB and C (see Figure 6). Patches G and B are parallel to
|T3 | > |T1 |, where |Ti | is the distance between two hinges on the ground and backdrop. Patches L and R are the left and
Ti (Figure 5b). right sides of the box structure, forming equal angles with
G and B, and are equidistant to the central fold. Patch C is a
3.2.3. V-fold special backbone glued perpendicularly to G and B at their
common fold line. On top of the box, patches TG and TB are
Our v-fold is also known as mechanism D1 in [LJGH11]. connected to L and R, respectively, and are glued to C at their
However, it may comprise more patches to depict more common fold line. In our work, all the fold lines connecting
shapes. In this work, we use two types of v-fold. B, R, TB , TG , L and G are made parallel to the central fold.
In the first type, when the primary patches are opened at On the front side of the box, patches FL and FR share a fold
180◦ , the v-fold structure forms a complete box as shown in and are connected to L and R, respectively. Similarly, BL and
Figure 5c. More specifically, our v-fold structure consists of BR are equivalent patches on the back of the structure. The
8 patches, G, B, FL , FR , BL , BR , TG , TB , in which we glue folds between FL , FR and between BL , BR must be copla-
{B, FR }, {G, FL }, {FL , TG }, {FR , TB }, and {FL , FR , BR , BL }. nar with patch C. Note that the front and back patches are
When B and G are opened at 180◦ , TG and TB are parallel to not glued to C. In principle, only one of the two pairs, either
B and G, while FL , FR , BL and BR are all perpendicular to B (FL , FR ) or (BL , BR ), is needed. However, in practice, box-
and G, and form 45◦ with the central fold. In addition, the style pop-ups normally contain both sides for better symme-
hinge between TG and TB lies in the bisecting plane of B and try and sturdiness. In addition, each side is allowed to have
G. Note that, in order for the top patches to fold up, we do multiple pairs of (FL , FR ) or (BL , BR ), as long as their fold
not glue {TB , BR } or {TG , BL }. lines do not intersect.

c 2014 The Author(s)
Computer Graphics Forum
c 2014 The Eurographics Association and John Wiley & Sons Ltd.
 C. R. Ruiz Jr., S. N. Le, J. Yu & K.-L. Low / Multi-style Paper Pop-up Designs from 3D Models

3.3. Pop-up Validity Based on the structures defined in Section 3.2, a tent-fold,
v-fold and box-fold can only be positioned along the central
As described in Section 3.1, a pop-up is said to be valid if
fold or between TG and TB of a box-fold. Step-fold can only
it is both foldable and stable. In this section, we discuss the
lie between a base patch and patch L or R of a box-fold. A
conditions for a valid pop-up.
base patch can be a primary patch, or TB , TG of a box-fold.
For instance, we can position a box-fold on top of another
3.3.1. Validity of Single-style Pop-ups
box-fold, then a step-fold between patch R of the upper box
Step-fold, tent-fold and v-fold pop-ups have been proven and TB of the lower box.
valid in earlier work [LJGH11, HS09]. We now show that
Note that a step-fold always forms parallel pairs of
box-fold is also valid.
patches, a tent-fold always remains symmetric to the central
hinge, and they do not move along the x-axis during folding.
Hence, we only need to consider the range of movement of
box-folds and v-folds for intersection checking.

Figure 7: Two cross sections of a box-fold.

Foldability of a box-fold. According to [HS09], if a 2D

quadrilateral with 4 sides a, b, c and d, in that order, has
a + b = c + d, then it can fold completely flat when a and c
are fully closed. Hence, a box with patches L, R, TG , TB , and
C is foldable if b + r = c + tb and g + l = c + tg . In addition, Figure 8: A fully-closed box-fold.
because the folds between FL , FR and between BL , BR are Box-fold. We consider a box-fold lying on patches B and
coplanar with the central fold and bisect the angle between G, in which the hinge between FL and FR is hF . Let d be
L and R, the front and back patches, FL , FR , BL , and BR , can the thickness of the box along the y-axis, and xF be the x-
also fold completely flat following the motions of L and R. coordinate of hF when the box is opened at 180◦ . Then when
Stability of a box-fold. We show that a box-fold as con- the box is fully closed (see Figure 8), the x-coordinate of hF
structed in Section 3.2.4 is stable. Assume that the structure becomes xF + d/2. Similarly, the x-coordinate of the hinge
is opened at an arbitrary angle in (0, 180◦ ]. If the angle be- hB on the back is xB − d/2 when the box is closed. Hence,
tween C and G may change while G and B are held station- to avoid intersection, no other mechanism should be placed
ary, then the box will undergo a shearing effect. However, it on B and G within the range [xB − d/2, xF + d/2] along the
also contains the front and back patches, FL , FR , BL , and BR , x-axis.
which keep L and R from shearing. In other words, C cannot
rotate while G and B are opened at an arbitrary angle. As a
result, a box-fold structure is stable.

3.3.2. Validity of Multi-style Pop-ups

Although each mechanism is foldable, the combined multi-
style pop-up may not be foldable, due to the possible inter-
sections between the mechanisms. In particular, during the
closing of a pop-up, patches FL , FR , BL , BR of a box-fold or
FL , FR of a v-fold emerge and may intersect with the corre-
sponding patches in another box-fold, v-fold, or tent-fold. To
make the whole pop-up foldable, it is important to position Figure 9: A fully-closed type-1 v-fold.
the mechanisms so that such intersections do not occur.
V-fold. In a type-1 v-fold, let xF and xB be the x-
First, we define a right-handed Cartesian coordinate sys- coordinates of the hinges between FL , FR , and between
tem, in which the z-axis is perpendicular to the central fold BL , BR when the fold is opened at 180◦ . Note that, when B
and bisecting the interior fold angle between the primary and G, the two base patches of the v-fold, are being closed,
patches. The x-axis lies along the central fold and the y-axis only FL and FR still touch G and B, and the v-fold leans to-
points in the direction of the ground patch. ward the positive direction of the x-axis. Hence, the small-

c 2014 The Author(s)

Computer Graphics Forum
c 2014 The Eurographics Association and John Wiley & Sons Ltd.
 C. R. Ruiz Jr., S. N. Le, J. Yu & K.-L. Low / Multi-style Paper Pop-up Designs from 3D Models

Figure 10: Overview of the automatic pop-up design algorithm.

est x-coordinate of the v-fold is xB . On the front of the v-

fold, the intersecting point between FL , FR , TB and√ TG has
the greatest x-coordinate, which becomes xF + l/ 2 at 0◦
fold angle, where l is the height of the v-fold along z-axis
when the fold angle is 180◦ . To avoid intersection, no other
mechanism √ should be placed on B and G within the range
[xB , xF + l/ 2] along the x-axis. The range for a type-2 v-
fold is similar, with xB being the x-coordinate of the right cut
lines of FL and FR when the fold is fully opened.
By following the two sufficient conditions above, we Figure 11: Model aligned with NPCA and the correspond-
avoid intersections between the considered mechanisms and ing bounding box.
guarantee that the combined pop-up is fully foldable. Fur-
thermore, because all the patches of each mechanism are sta- are connected to form the central fold line, where the sym-
ble, their resulting multi-style structure is also stable. metry plane intersects the base.
We then perform a RANSAC primitive fitting similar to
4. Automatic Pop-up Design Algorithm the techniques used for point clouds [SWK07]. However,
in our case we fit volumes rather than surfaces, similar to
Our algorithm takes a 3D mesh as input, then generates a 2D the work of [YK12]. Since some meshes may only have a
design layout of a valid paper pop-up. Our main idea is to few sparse vertices, we also employ a preprocessing step to
abstract a mesh into sub-volumes by fitting 3D primitives, add pseudo vertices to the input mesh in a technique simi-
and then choose the best mechanism to represent each prim- lar to [SLCH11] for grid-based PCA. This ensures that our
itive. Each primitive-mechanism pair has its own set of steps vertices are evenly distributed on the object’s surface.
to convert a shape into a set of valid pop-up patches that is
guaranteed by our formulation. Once the patches have been Our primitives also have some constraints on their ori-
generated and stabilized, we produce a design layout and de- entation and connectivity. We primarily have two basic 3D
termine an assembly order for the printable instructions. primitives, which are the rectangular and triangular prisms.
Because of the constraints on these primitives, we only need
a minimum of two points to specify a rectangular prism and
4.1. 3D Primitive Fitting four for a triangular prism. In these prisms, the sides edges
First, we align the input model using Normal Principal Com- are always parallel to one another, while the top and bottom
ponent Analysis (NPCA). We obtain the principal axis of the edges form a rectangle or a triangle.
3D model using the surface normals weighted by the area of An important part of the RANSAC primitive fitting is the
the faces. Our aim is to automatically align the model such error metric used to judge how well the fitting is. The error
that the principal axis will point in the same direction as the of a set of points to a primitive is given by:
normal of the faces with the most cumulative surface area.
We set this principal axis as our y-axis, and the x- and z-axis
error f it = outliers ∗ w1 + dist ∗ w2 + coverage ∗ w3 (1)
to the vectors orthogonal to this axis. It is also possible to al-
low the user to align the model based on his own preference.
where outliers is the percentage of points that are not part
Alternatively, we can also use the dominant long edges in the
of the consensus set of the current model, dist is the cumula-
model to determine the alignment to facilitate foldability.
tive normalized distance of the points to their nearest surface
Then we obtain the symmetry plane and the tightest axis- on the primitive, and coverage is the volume covered by the
aligned bounding box of the model. The base of the bound- primitive relative to the bounding box of all the points. The
ing box forms our initial backdrop and ground patches. They weights w1 , w2 and w3 are in the range from 0 to 1 and add

c 2014 The Author(s)
Computer Graphics Forum
c 2014 The Eurographics Association and John Wiley & Sons Ltd.
 C. R. Ruiz Jr., S. N. Le, J. Yu & K.-L. Low / Multi-style Paper Pop-up Designs from 3D Models

up to 1. They control the influence of each metric to the total RANSAC is used to fit the vertices to individual primi-
error. These weights affect the number of primitives fitted by tives. In order to get the best combination of primitives that
the RANSAC algorithm. By putting more weight on cover- satisfy the aforementioned constraints, we formulate it as a
age and outliers, it will favor bigger volumes that encompass combinatorial optimization problem. We then use dynamic
more vertices and in turn produce bigger and fewer patches. programming to minimize the accumulated individual prim-
itive errors (error f it ).
The spatial constraints on our primitives are:
Constraint 1. Angles: Every angle on the rectangular
prism must be 90◦ . Similarly, the angle between a top or 4.2. Mechanism Mapping and Primitive Refitting
bottom edge and a side edge in a triangular prism must also
be 90◦ . After the 3D primitive fitting step, we now have a 3D model
that is abstracted using primitives that best approximate the
Constraint 2. Orientation: The rectangular prism’s four original shape and at the same time can be mapped to pop-
side edges should only be either parallel or orthogonal to the up mechanisms. Using the valid primitive-mechanism pairs
central fold line (x-axis). The triangular prism’s three side shown in Figure 3, we derive Table 1.
edges, must similarly be either parallel or orthogonal to the
central fold line (x-axis).
Primitive Step V-fold Box Tent
Rectangular Prism X X X
Triangular Prism X X
Table 1: Possible primitive-to-mechanism mappings.

Most of the abstraction and approximation is done in the

previous step. Normally, there will only be a few possible
mappings of the primitives and we can do an exhaustive
search of all the possibilities. We select the combination
Figure 12: Valid orientations of the 3D primitives. that minimizes a certain error criterion. Our error criterion
is based on the coverage of the mechanism or how well its
Constraint 3. Placement: A new primitive can only be
patches can approximate the primitive:
placed on a Base Patch Pair. A Base Patch Pair (BPP) is
a set of two patches that share a common hinge and can
act as intermediary backdrop and ground patches. For ex- errorcover = 1 − (∑ Area principal f ace / ∑ Area f ace ) (2)
ample, in Figure 13, we start with our first BPP, which are
the primary patches {B, G}. Once we add the rectangular Unlike [LJGH11] that uses all the faces of the voxel as
prism in the center, three more base pairs are generated: patches, we only use some faces of the primitives called the
{F1, F2}, {F3, B}, {F4, G}. Adding another rectangular principal faces (the shaded faces in Figure 3). The error is
prism on {F4, G} generates two more base pairs {F4, F5} based on the surface area covered by only the principal faces
and {F6, G}. Finally, we add a triangular prism on {F1, F2} over all the surface area of the faces of the primitives in the
pair, and this does not add any new BPP. structure.
During the mechanism mapping, we also examine the va-
lidity of the generated structure based on the conditions in
Sections 3.3.1 and 3.3.2. These formulations allow us to test
the foldability and stability of the structure without the need
for simulations.
The primitives can also be refitted as long as they sat-
isfy the foldability of the mechanisms. For example, the side
patches for box-folds do not necessarily need to be orthogo-
nal to the ground or backdrop patches, as long as the lengths
Figure 13: Base Patch Pairs. BPP = {{B, G}, {F1, F2},
of the patches still satisfy b + r = c + tb and g + l = c + tg ,
{F3, B}, {F4, G}, {F4, F5}, {F6, G}}.
such as the case of the ship in Figures 14 and 15, where
These constraints are based on the possible combinations the sides of the box are refitted to slant along the body of
of the different mechanisms, as described in Section 3.3.2. the ship. In addition, if sufficient gap between different box-
For example, a triangular prism is permitted on top of a rect- folds along a common hinge is not achievable, the front and
angular prism, but not on its side, since such combination back patches of the box-folds can be partially trimmed to
may not map to a foldable pop-up. avoid the intersection.

c 2014 The Author(s)

Computer Graphics Forum
c 2014 The Eurographics Association and John Wiley & Sons Ltd.
 C. R. Ruiz Jr., S. N. Le, J. Yu & K.-L. Low / Multi-style Paper Pop-up Designs from 3D Models

5. Results
We run our algorithm on an Intel Core i7 PC with 8GB of
RAM. The entire process completes in a few seconds to a
few minutes with the primitive fitting accounting for most
of the running time. We have tested our approach on several
3D models from the Google 3D Warehouse [Goo13], where
Figure 14: Refitting a rectangular prism. some models contain up to 30000 vertices. Figure 15 shows
4.3. Patch Generation some of the models we have used.
Once we know the mapping of the primitives, we now con- Our approach works well for block-like models, such as
vert each principal face of the primitive into a pop-up patch. man-made objects (e.g. cars, trucks, buildings), since they
Unlike previous approaches that simply consider the entire can be closely approximated using our set of primitives. Al-
face as a patch, we use an image-based approach to better though our primitives are symmetric, the combinations of
approximate the shape and contours of the object. In gen- these primitives can be asymmetric structures. Organic and
eral, we render an orthographic projection of the mesh unto rounded models might require more complex primitives and
a plane co-planar with the principal face. This produces an higher levels of abstraction (see Figure 17). Objects that are
image that becomes the basis of the shape and texture of the hard to align to the principal axes may also be difficult to
patch. approximate exactly. In some cases, like the T-shape, neces-
sary additional supports may also distract from the original
Box-folds are of particular interest because the front, back
shape.
and top patches (FL , FR , BL , BR , TG , TB ) can be composed
of multiple layers of patches, which can better approximate We compare our results with those of [LSH∗ 10] and
the volume of the original mesh. For example, the box-folds [LJGH11] in Figure 16. Because of the voxelization, their
used in the car and monster truck in Figure 15 are composed results contain many small box structures, which are diffi-
of several front, back and top patches at different depths or cult to make in practice. Our pop-ups consist of only two
heights. These patches are obtained by performing image box-folds for the Waldorf-Astoria Hotel and a single tent-
segmentation on the depth and normal maps taken from an fold for the Eiffel Tower. Although our geometry is not as
orthographic projection of the front, back and top view. detailed, we are able to use textures to capture the details
and shape information.
Basically, the image segmentation works by locally fitting
a quadratic surface on the segmented pixels in the neighbor- Most of our pop-up designs employ only a few mecha-
hood of a candidate pixel, similar to [LLLN∗ 13]. It deter- nisms and as such only a few patches. This is intentional
mines whether a pixel p should be part of the current seg- since our main goal is to generate the simplest design that is
ment by thresholding f (p)−q(p), where f (p) is the depth still able to capture the general shape of the input 3D model.
value and the x-, y-, z-components of the normal vector at p, We have determined that in most cases capturing the side
and q(p) is the quadratic approximation from the previously contours of a 3D model is enough to give us a good abstrac-
segmented pixel p0 : tion of the model.
1 00
q(p) = f (p0 ) + f 0 (p0 )(p − p0 ) + f (p0 )(p − p0 )2 (3)
2

4.4. Design Layout Generation

After we obtain the patches, we continue to generate the de-
sign layout for cutting, folding and gluing. We position the
patches separately on a sheet of paper, adding a flap to each
edge where it connects with another patch. When two ad-
jacent patches are bounded by cut lines and their common
hinge, we connect them in the layout so that the user will
only need to create a fold without gluing. We finally assign
IDs to the patches and flaps accordingly.
The order of assembling the patches is important. Some
orders are not feasible, while some are easier to construct
in practice. In our work, we employ a method similar to
[APH∗ 03], which produces assembly instructions for rigid
components. While we achieve feasible assembly orders for Figure 17: Our actual paper pop-ups for Stanford Bunny
our pop-ups, further study is needed to improve the ease of (textured using the rendered model), skewed cube, half-
assembly. sphere and T-shape.

c 2014 The Author(s)
Computer Graphics Forum
c 2014 The Eurographics Association and John Wiley & Sons Ltd.
 C. R. Ruiz Jr., S. N. Le, J. Yu & K.-L. Low / Multi-style Paper Pop-up Designs from 3D Models

Figure 15: Sample results. Input models and their corresponding actual pop-ups.

Figure 16: Waldorf-Astoria Hotel and Eiffel Tower models. (a) Input 3D model, (b) [LSH∗ 10] results, (c) [LJGH11] results
(from paper) and (d) our actual paper pop-ups.

6. Conclusions and Future Work Our approach has difficulty in approximating curved and
rounded parts, since we only have rectangular and triangu-
We present an original algorithm to generate multi-style pop-
lar prisms as primitives. Professional artist are able to ap-
up designs automatically. Some of the individual steps (e.g.
proximate these shapes in their work using very specialized
primitive fitting) are well-known graphics techniques, but
rounded or curved mechanisms. These specific mechanisms
their combination and their use in similar context, with ex-
however are not included because their formulations would
plicit consideration of pop-up validity, have not been done
require a different definition of stability, under the assump-
in previous work. We believe our approach provides a novel
tion of non-rigid and bendable paper.
and reasonable framework for combining multiple pop-up
mechanisms.

c 2014 The Author(s)

Computer Graphics Forum
c 2014 The Eurographics Association and John Wiley & Sons Ltd.
 C. R. Ruiz Jr., S. N. Le, J. Yu & K.-L. Low / Multi-style Paper Pop-up Designs from 3D Models

Our system is currently fully automatic. However, it may [HS09] H ARA T., S UGIHARA K.: Computer-aided design of pop-
be important for the user to be given creative control over up books with two- dimensional v-fold structures. In Proc. 7th
the final design. We may achieve that in the future by incor- Japan Conf. on Comput. Geometry and Graphs (2009). 4, 5
porating our approach into other interactive systems, such [Hul06] H ULL T.: Project Origami: Activities for Exploring
as those by [Gla02a, Gla02b, HE06]. In addition, the pre- Mathematics. A K Peters, Ltd, 2006. 2
sented geometric formulations here do not take into account [IEM∗ 11] I IZUKA S., E NDO Y., M ITANI J., K ANAMORI Y.,
F UKUI Y.: An interactive design system for pop-up cards with a
the physical characteristics of paper. In actual pop-up de-
physical simulation. Vis. Comput. 27, 6 (June 2011), 605–612. 3
sign, the thickness, mass, strength and elasticity of paper are
important considerations. Lastly, we do not have quantita- [LJGH11] L I X.-Y., J U T., G U Y., H U S.-M.: A geometric study
of v-style pop-ups: theories and algorithms. In ACM SIGGRAPH
tive assessment of the aesthetic quality of our pop-ups. Such 2011 (2011), pp. 98:1–98:10. 2, 3, 4, 5, 7, 8, 9
measurements can be very beneficial in creating more visu-
[LLLN∗ 13] L E S. N., L EOW S.-J., L E -N GUYEN T.-V., RUIZ
ally appealing paper pop-up designs. C., L OW K.-L.: Surface- and contour-preserving origamic archi-
tecture paper pop-ups. IEEE Trans Vis Comput Graph (TVCG)
(Aug 2013). DOI:10.1109/TVCG.2013.108. 2, 3, 8
Acknowledgements
[LNLRL13] L E -N GUYEN T.-V., L OW K.-L., RUIZ C., L E .
This work is supported by the Singapore MOE Academic S. N.: Automatic paper sliceform design from 3d solid models.
Research Fund (Project No. T1-251RES1104). We would Trans Vis Comput Graph (Nov 2013). 2
also like to thank Armandarius Darmadji for his help in the [LSH∗ 10] L I X.-Y., S HEN C.-H., H UANG S.-S., J U T., H U S.-
implementation. M.: Popup: automatic paper architectures from 3d models. In
ACM SIGGRAPH 2010 (2010), pp. 1–9. 2, 3, 4, 8, 9
[LTS96] L EE Y., T OR S., S OO E.: Mathematical modelling and
References simulation of pop-up books. Comp. & Graph. 20, 1 (1996), 21–
31. 3
[ADD∗ 12] A BEL Z., D EMAINE E. D., D EMAINE M. L., E ISEN -
STAT S., L UBIW A., S CHULZ A., S OUVAINE D. L., V IGLI - [MS04] M ITANI J., S UZUKI H.: Making papercraft toys from
ETTA G., W INSLOW A.: Universality results for pop-up cards. meshes using strip-based approximate unfolding. ACM Trans.
Preprint, 2012. 3 Graph. 23, 3 (2004), 259–263. 2
[APH∗ 03] AGRAWALA M., P HAN D., H EISER J., H AYMAKER [MSM11] M C C RAE J., S INGH K., M ITRA N. J.: Slices: a shape-
J., K LINGNER J., H ANRAHAN P., T VERSKY B.: Designing ef- proxy based on planar sections. ACM Trans. Graph. 30, 6 (2011),
fective step-by-step assembly instructions. ACM Trans. Graph. 168:1–168:12. 2
22, 3 (July 2003), 828–837. 8 [MSU03] M ITANI J., S UZUKI H., U NO H.: Computer aided de-
[BH02] B ELCASTRO S.-M., H ULL T. C.: Modelling the fold- sign for origamic architecture models with voxel data structure.
ing of paper into three dimensions using affine transformations. Trans. of Inf. Process. Soc. of Japan 44, 5 (2003), 1372–1379. 3
Linear Algebra and its Applications 348 (2002), 273–282. 4 [MWZ∗ 13] M ITRA N., WAND M., Z HANG H., C OHEN -O R D.,
[Car08] C ARTER D. A.: Yellow Square: A Pop-Up Book for Chil- B OKELOH M.: Structure-aware shape processing. In Eurograph-
dren of All Ages. Little Simon, 2008. 1, 2 ics State-of-the-art Report (STAR) (2013). 3
[Cro11] C ROWTHER R.: Amazing Pop-Up Monster Trucks. [MZL∗ 09] M EHRA R., Z HOU Q., L ONG J., S HEFFER A.,
Walker & Company, 2011. 1, 2, 4 G OOCH A., M ITRA N. J.: Abstraction of man-made shapes. In
[CS03] C ARROLL L., S ABUDA R.: Alice’s Adventures in Won- ACM SIGGRAPH Asia 2009 papers (2009), pp. 137:1–137:10. 2
derland: A Pop-up Adaptation. Little Simon, 2003. 1, 2 [O’R11] O’ROURKE J.: How to Fold It: The Mathematics of
[DDSD03] D ÉCORET X., D URAND F., S ILLION F. X., D ORSEY Linkages, Origami & Polyhedra. Cambridge Uni. Press, 2011.
J.: Billboard clouds for extreme model simplification. In ACM 2
SIGGRAPH 2003 (2003), pp. 689–696. 2 [SLCH11] S HIH J.-L., L EE C.-H., C HAO -H UNG C.: A 3d
[DO07] D EMAINE E., O’ROURKE J.: Geometric Folding Algo- model retrieval approach based on the combination of pca plane
rithms: Linkages, Origami, Polyhedra. Cambridge Uni. Press, projections. J. Info. Tech. and Appl. 5, 2 (2011). 6
2007. 2 [SWK07] S CHNABEL R., WAHL R., K LEIN R.: Efficient ransac
[Gla02a] G LASSNER A.: Interactive pop-up card design, part 1. for point-cloud shape detection. Comput Graph Forum 26, 2
In IEEE Comput. Graph. Appl. (2002), vol. 22, pp. 79–86. 3, 10 (June 2007), 214–226. 6

[Gla02b] G LASSNER A.: Interactive pop-up card design, part 2. [Tac10] TACHI T.: Origamizing polyhedral surfaces. IEEE Trans.
In IEEE Comput. Graph. Appl. (2002), vol. 22, pp. 74–85. 3, 10 Vis. Comput. Graphics 16, 2 (2010), 298–311. 2

[Goo13] G OOGLE:. http://sketchup.google.com/3dwarehouse, [UT06] U EHARA R., T ERAMOTO S.: The complexity of a pop-
2013. 8 up book. In 18th Canadian Conf. on Comput. Geom. (2006). 3

[HE06] H ENDRIX S. L., E ISENBERG M. A.: Computer-assisted [vKXZ∗ 13] VAN K AICK O., X U K., Z HANG H., WANG Y., S UN
pop-up design for children: computationally enriched paper en- S., S HAMIR A., C OHEN -O R D.: Co-hierarchical analysis of
gineering. Adv. Technol. Learn. 3, 2 (2006), 119–127. 3, 10 shape structures. ACM Transactions on Graphics (Special Issue
of SIGGRAPH) 32, 4 (2013). 3
[HEH05] H OIEM D., E FROS A. A., H EBERT M.: Automatic
photo pop-up. In ACM SIGGRAPH 2005 (2005), pp. 577–584. 3 [YK12] Y UMER M. E., K ARA L. B.: Co-abstraction of shape
collections. ACM Trans. Graph. 31, 6 (Nov. 2012), 166:1–
[HEH08] H OIEM D., E FROS A. A., H EBERT M.: Closing the 166:11. 2, 6
loop in scene interpretation. In IEEE CVPR (2008). 3

c 2014 The Author(s)
Computer Graphics Forum
c 2014 The Eurographics Association and John Wiley & Sons Ltd.
180° Unequal Tent Fold
180˚ Box with a Parallel Plane
Figure 13