A Non-Photorealistic Rendering Technique for Art-directed Hatching of

3D Point Clouds

Ronja Wagner, Ole Wegen

, Daniel Limberger

, J

urgen D

ollner and Matthias Trapp

Hasso Plattner Institute, Faculty of Digital Engineering, University of Potsdam, Germany

Keywords:

3D Point Clouds, Non-photorealistic Rendering, Real-time Rendering, Art-directed.

Abstract:

Point clouds or point-based geometry of varying density can nowadays be easily acquired using

LiDAR

cameras

or modern smartphones with

LiDAR

sensors. We demonstrate how this data can be used directly to create novel

artistic digital content using Non-Photorealistic Rendering techniques. We introduce a

GPU

-based technique

for art-directable

NPR

rendering of 3D point clouds at interactive frame-rates. The technique uses either a

subset or all of the points to generate oriented, sketchy strokes by taking local curvature and normal information

into account. It uses X-Toon textures as part of its parameterization, supports hatching and cross hatching, and

is inherently temporal coherent with respect to virtual camera movements. This introduces signiﬁcant artistic

freedom that is underlined by our results, which show that a variety of different sketchy styles such as colored

crayons, pencil, pointillism, wax crayons, blue print, and chalk-drawings can be achieved on a wide spectrum

of point clouds, i.e., covering 3D polygonal meshes as well as iPad-based LiDAR scans.

1 INTRODUCTION

Point clouds are a simple, compact, and ﬂexible geo-

metric representation that can easily be acquired for

real-world scenes using off-the-shelf photogrammet-

ric techniques. With the introduction of Light Detec-

tion And Ranging (

LiDAR

) sensors in high-end smart-

phones, e.g., iPhone 12, straightforward acquisition of

point clouds is easily accessible to consumers.

Despite both its potential applicability in various

domains and increasing popularity in social media—

mainly for 3D photography and animation with the

point cloud as an art style in itself— their usage

as a fundamental geometric representation in Non-

Photorealistic Rendering (

NPR

) as well as respective

rendering systems or frameworks are sparsely stud-

ied. Notable uses of 3D point clouds are (1) to enable

light transport for rendering scanned assets (Sabbadin

et al., 2019), (2) stylized, interactive self portraits, (3)

data visualization, as well as (4) art-directed particle

animation and stylization such as the reprojection of

recorded memories in the computer game Cyberpunk

2077 (Ankermann et al., 2021) as a means of distin-

guishing between reality and virtual reality. Another

noteworthy use can be found in previsualization or

https://orcid.org/0000-0002-6571-5897

https://orcid.org/0000-0002-9111-4809

https://orcid.org/0000-0003-3861-5759

(a) Blue Print (b) Pencil

Figure 1: The Stanford dragon as 3D point cloud, rendered

in four different styles using art-direct hatching.

digital content creation to evaluate ideas and concepts

early before performing possibly costly data enhance-

ment or transformation, e.g., creating stylized, textured

3D meshes. We present a ﬁrst approach into this do-

main by researching methods to enable art-directed

procedural-controlled rendering effects.

220

Wagner, R., Wegen, O., Limberger, D., Döllner, J. and Trapp, M.

A Non-Photorealistic Rendering Technique for Art-directed Hatching of 3D Point Clouds.

DOI: 10.5220/0010849500003124

In Proceedings of the 17th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2022) - Volume 1: GRAPP, pages

220-227

ISBN: 978-989-758-555-5; ISSN: 2184-4321

In the context of

NPR

, point clouds have both ad-

vantages and disadvantages over traditional 3D polyg-

onal models. Point clouds lack inherent connectiv-

ity information that needs to be respected or main-

tained, which simpliﬁes their representation, transfor-

mation, and rendering. The absence of surface-related

attributes such as normal vectors, texture coordinates,

or curvature data is rather inconvenient, since they are

essential for the implementation of

NPR

techniques.

Additionally, while high quality renderings of point

clouds are possible (Sch

utz and Wimmer, 2015; Sch

utz

et al., 2021), they often rely on a sufﬁcient point den-

sity. With these observations in mind, enabling the

implementation of real-time, art-directed

NPR

tech-

niques comprises the following major challenges:

Art-directable Parametrization (C

Decoupling

of processing operations that are required for

different manifestations of stylization techniques,

e.g., point-splat functions (Anjos et al., 2018), to

facilitate the interchange of respective techniques

and increasing ease-of-use.

Efﬁcient Data Handling (C

An efﬁcient represen-

tation of complex point clouds and respective styl-

ization data to allow for art-directed stylization

based on point-attribute data. This also concerns

the reduction of data transfer and update latency to

support real-time rendering for interactive control.

Existing approaches to implement animation tech-

niques for point clouds are (1) customized integration

into existing game engines (to enable real-time render-

ing) or (2) speciﬁc tooling or scripts to extend existing

software, e.g., Blender. These, however, either re-

quire auxiliary constructs for data representation, e.g.,

texture-based representations, lack real-time rendering

capabilities, or can hardly handle millions of points.

In the context of the challenges

and

, we

present a Graphics Processing Unit (

GPU

)-based

hatching technique for producing art-directed non-

photorealistic renderings of potentially massive point

clouds, which are obtained either using

LiDAR

scan-

ning of a real-world scene or by sampling polygonal

geometry. The presented approach is not limited to

hatching, but can serve as a general approach for the

implementation of further non-photorealistic rendering

techniques for point clouds.

To this end, we. . .

. . . introduce a method for creating sketched im-

ages directly out of 3D point clouds using only

position, normal, and curvature information,

enable signiﬁcant artistic freedom due to wide

range of parameters by using X-Toon textures, and

demonstrate the method’s real-time capability (pre-

computed attributes) and temporal coherence.

2 RELATED WORK

Prior art with respect to texture-based stylization in 3D

object space using, e.g., proxy-geometry as well as in

image space remains quite popular and is well covered.

In the following, we brieﬂy classify an exemplary cross

section of classics as well as latest works.

NPR

rendering of 3D polygonal meshes targeting

a sketched look and feel can be easily achieved us-

ing texture-based hatching approaches (Praun et al.,

2001; Webb et al., 2002). Hatching patterns are pre-

generated in multiple levels of details, i.e., Tonal Art

Maps, and enable ﬁne control over shading without

aliasing. In order to also create sketchy outlines

or silhouettes, image-based approaches can be ap-

plied (Nienhaus and D

ollner, 2003; Zakaria and Seidel,

2004). These types of techniques, however, cannot be

applied to point clouds, since texturing and image-

based post processing is implemented fundamentally

different; Strokes are placed directly and individually,

and require a dedicated level of detail concept. In ad-

dition, image-based curvature estimates (Kim et al.,

2008) and purely image-based stippling (Awano et al.,

2010) are not applicable and, regardless, usually lead

to temporal incoherent results.

Another common approach for simulating, e.g.,

pencil-like renderings, is to use edges of the 3D meshes

to create overlapping graphite strokes (Sousa and

Buchanan, 1999). This concept of creating textured

proxy geometry, also sometimes referred to as graph-

tals (Smith, 1984), facilitates temporal frame-to-frame

coherence and can be applied to particles and point

clouds alike (Kaplan et al., 2000; Markosian et al.,

2000). The proxy geometry, e.g., dots or strokes, can

be specialized dynamically at run-time w.r.t. size, ori-

entation, placement, and color. Thereby, artistic con-

trol can be linked to mesh properties and thus allow

for silhouettes and contouring (Grabli et al., 2004).

This approach represents the core of our technique

and, apart from various differences in the implementa-

tion details due to the evolution of graphic hardware,

can be easily extend, enhanced, and specialized for

point clouds.

A promising extension to this are ribbons, i.e., one-

dimensional strips of points, ordered “according to

their neighbourhood relations” (Runions et al., 2007).

The concept could be used to create silhouette or cur-

vature aware proxy geometry (Schmid et al., 2011)

such as strokes consisting of multiple links instead of

single, straight ones, and might allow for ﬁne-grained

stippling styles. So far, existing approaches covering

water color, caligraphy, etc. still rely on strokes drawn

by the artist (Zheng et al., 2017). Automatic splatting

of lines or strokes, though applicable to volumetric

A Non-Photorealistic Rendering Technique for Art-directed Hatching of 3D Point Clouds

221

(a) Pencil. (b) Wax crayon. (c) Two-colored.

Figure 2: Shading effects and their corresponding X-Toon

textures (RGBA). Note that the y-axis is not utilized here.

data as well, does only cover a speciﬁc style and is not

easily art-directable (Zhang et al., 2014; Anjos et al.,

2018). We partially combine these ideas and aim for

an automated approach that allows for a constrained

but artist-directable, expressive stylization.

3 ART-DIRECTED HATCHING

This section details concepts as well as implementation

details of our real-time hatching technique.

3.1 Art-directed Parametrization

We make the assumption that any point cloud is given

with point normal vectors (Cao et al., 2021), color, and

curvature attributes (M

erigot et al., 2009; Lu et al.,

2020). The latter two can improve the appearance but

are not required. To achieve the desired hatching effect,

each point is replaced by textured proxy geometry, e.g.,

two orthogonal lines aligned to the surface, textured

with pencil strokes.

A drawing technique that characterizes cross-

hatching is to depict lighting by not only using dif-

ferent colors but also by omitting lines and therefore

showing the underlying paper. The amount of light

that reaches a point deﬁnes how likely it is to be omit-

ted. To accomplish a wide variety of artistic styles and

effects, the color and point density for each amount

of light can be controlled through the alpha channel

of an X-Toon (Barla et al., 2006; Kang et al., 2009)

texture (Fig. 2). On white paper for example, strongly

illuminated parts are realized by removing most points,

while darker parts keep their original point density. An-

other option (better suited to darker backgrounds) is

to color well-lit points brightly and leave out points

receiving a medium amount of light.

Lighting by omission only works if there are no

visible parts behind the points that are left out. To

avoid such parts, e.g., the back of the model, from

appearing where only the paper should be visible, we

use an additional depth pass. With that pass we ensure

that nothing behind the surface of the model is visible

in the ﬁnal result.

For a more realistic pencil effect there is also the

option to limit the stroke colors to a number of pre-

deﬁned pencil colors, similar to a colored pencil set.

In addition to the classic X-Toon parametrization, the

following parameters are provided for hatching:

Line Length and Thickness:

These parameters con-

trol the extent of hatching strokes.

Parallel to cross-hatch Proportion:

Our technique

supports both cross hatching, where almost all

points are replaced with two lines, and parallel

hatching, where all points are replaced with a sin-

gle line. This parameter speciﬁes the probability

for a point to be replaced with a single stroke or

two, crossed strokes.

Stroke Density:

In point clouds of high density, the

individual hatched strokes might not be distinguish-

able from each other. By specifying the proportion

of strokes to be left out, the density of rendered

strokes can be controlled.

Line Direction:

This parameter is used to determine

the direction of each stroke.

Light Direction:

For lighting we use a uniform light

direction that is speciﬁed by this parameter.

Coloring Mode:

Deﬁnes which method is to be used

for determining the color of the hatched strokes.

Hue Shift:

Is applied to the color calculated with the

selected coloring mode.

3.2 Stroke Generation for Hatching

X-Toon Coordinate Computation.

To determine

the desired density and therefore the proportion of

points to be omitted, we ﬁrst compute the X-Toon

texture coordinates, namely depth and brightness, for

a point. Transforming the point’s z-coordinate (af-

ter projection) into a

[0,1]

range yields the depth co-

ordinate. The brightness coordinate depends on the

scene lighting implementation, e.g., Phong-based light-

ing (Phong, 1975). With these coordinates and the

corresponding texture value, we determine whether

the current point should be discarded or replaced by

lines, and in the latter case, obtain the color of these

lines.

Probabilistic Interpolation.

Both lighting by omis-

sion and the pencil color quantization rely on a proba-

bilistic approach. Instead of interpolating between two

values (e.g., visible and not visible or two RGB val-

ues) and applying the result, the point only attains one

of the values chosen at random with the interpolation

factor determining the probabilities. With a sufﬁcient

number of points, the result creates an approximate

GRAPP 2022 - 17th International Conference on Computer Graphics Theory and Applications

222

(b) Irregular Stroke Placement(a) Regular Stroke Placement

Figure 3: For hatching, one or more strokes, i.e., two for

cross-hatching, can be placed and oriented either in a (a) reg-

ular more accurate or (b) irregular and scattered way.

impression of directly interpolated values. As random

seed, we use the point ID, which is temporally consis-

tent. For determining whether a point should be visible

or not, we use the alpha value from the X-Toon texture

as the interpolation factor between visibility and invis-

ibility. So if

a ∈ [0;1]

is the alpha value, the point has

100 · a%

chance of being visible. The general point

density can be controlled through a corresponding pa-

rameter that we multiply with the alpha value from the

X-Toon texture.

Line Geometry Generation.

For the cross hatch-

ing effect we replace each point with two orthogonal

lines perpendicular to the point’s normal. In doing so,

we need to ensure the line directions are consistent

for all points. Line directions are consistent if points

with similar normal vectors produce similar line direc-

tions (Fig. 3) so that they properly illustrate the surface

direction. To achieve that consistency we project a nor-

malized uniform vector on the plane deﬁned by each

point normal. Let

be the uniform vector and

be the

surface normal. The projected vector

is computed

as follows:

= ~u − (~u ·~n) ·~n

. The second direction

can be determined by simply computing the nor-

malized cross product of the ﬁrst direction and the

point normal:

=~n ×

. With these directions, we

can now create two orthogonal lines along the surface.

Let

be the point coordinates and

be the line length

parameter. Then the ﬁrst line extends from

p −

p +

. The second line starts at

p −

and ends at

p +

. We can now use these coordinates to create

rectangle geometry for each line. One side of said rect-

angle is described by the corresponding line direction

and length. The direction of the other side needs to

be orthogonal to both the line direction and the vector

from the camera to the current point and can therefore

be calculated by taking their cross product. The width

of the line is deﬁned in a separate line thickness pa-

rameter. From this information, we can derive the four

vertices of the resulting rectangle, with the texture co-

ordinates

(0,0)

(0,1)

(1,1)

and

(1,0)

for the pencil

stroke texture.

3.3 Local Shading Models

There are three ways to determine the color of the

created pencil strokes. The simplest one is to apply

the same color to all strokes, which can, for example,

be used to create a classic gray pencil sketch or a blue

ballpoint pen drawing. Alternatively, the color can

be taken directly from the X-Toon texture. The third

approach is to blend the initial point color with the

X-Toon texture value to depict both lighting and the

original color information.

After the point color is determined, it is possible to

add pencil color quantization. That means only allow-

ing colors from a predeﬁned set to simulate drawing

with pencils from a limited pencil box. Unlike other

media, pencil colors can only be mixed to a limited

extent. With watercolors for example, three colors and

black and white are in theory sufﬁcient to get every

other color. For colored pencils however, you need

more base colors to illustrate the whole color spec-

trum properly. This is because generally, the quality

of the blending result of two pencil colors suffers with

decreasing similarity of the blended colors.

Our pencil color effect simulates that by calcu-

lating the two closest colors from a predetermined

pencil set to a given color and blending them with

probabilistic interpolation. We call the given color

. The pencil color set is given as rgb and hsv val-

ues sorted by hue. It contains only colors with high

saturation and value. First, we calculate the hsv rep-

resentation of

and apply the hue shift parameter.

We can then search for the colors

and

from the

pencil set whose hues are closest to

’s hue, so that

hue(c

) = hue(c

)· a + hue(c

)· (1 − a)

applies for an

a ∈ [0,1]

. We can use this

as an interpolation factor

for probabilistic interpolation to create the impression

of a color with the same hue as

. Finally, we mix the

resulting color with black and white to more closely

match c

in value and saturation.

(a) (b) (c)

Figure 4: The point sprites used during the depth pass

cause discontinuities in the z-buffer which interfere with the

strokes rendered later (a). In addition, the semi-transparent

strokes interfere with one another (b). By shifting the strokes

towards the virtual camera using polygon offsets while using

order-independent rendering (blending or alpha to coverage)

both effects can be circumvented respectively (c).

A Non-Photorealistic Rendering Technique for Art-directed Hatching of 3D Point Clouds

223

3.4 Depth Pass

For the lighting by omission approach to work, we

need to ensure there are no visible lines behind the

points that are omitted. A ﬁrst approach to that prob-

lem is back face culling, which prevents the back of the

surface from being visible. However, even with back

face culling, there still are lines behind the surface that

are not supposed to be visible (Fig. 6). We prevent

that by using an additional rendering pass before the

hatching pass.

In the depth pass, we render every point as a single

circle and write only to the depth buffer. With a sufﬁ-

cient point density and point size, these circles cover

the whole surface, so in the following hatching pass,

only lines in front of the surface get rendered. Without

any modiﬁcations however, the circles of the points at

the front will cover the pencil strokes of nearby points.

To avoid that problem, we add an offset to the poly-

gons in the hatching pass, so the pencil strokes are

slightly in front of their corresponding points from the

depth pass (Fig. 4 a). Since the depth test needs to be

activated in the drawing pass for this approach to work,

we need to make sure nothing is written to the depth

buffer in the hatching pass, otherwise the lines will

cover each other instead of overlapping and thereby

creating darker tones (Fig. 4 b).

The size of the circles is also controllable via a pa-

rameter and needs to be adapted to achieve the desired

result in the depth pass and to not cover too much or

to little of the hatched lines.

3.5 Enhancements

The approach described so far allows for decent results,

though the following tweaks should be considered.

Alpha Modiﬁcation.

So far, we use the approach

of probabilistic interpolation to create the impression

of the surface being transparent. However, for image

regions with almost full transparency, this results in

very few strong strokes on an otherwise clear part

of the surface, when in an actual pencil sketch, you

would also adjust the pressure of the pencil to modify

the transparency of the strokes themselves. That is

(a) Points (b) W/o alignment (c) With alignment

Figure 5: The length of strokes can be aligned based on the

local curvature at the stroke’s location. This can be used to

achieve (b) increased sketchiness especially in curved areas

as well as (c) a strict, more accurate hatching.

why we adjust the alpha value of the generated pencil

stroke proportional to the alpha value from the X-Toon

texture. Also, even in parts with similar lighting, not

every stroke is applied with exactly the same pressure.

To account for that, we apply some randomness on the

ﬁnal alpha value.

Interpolation of Cross & Parallel Hatches.

In ad-

dition to a decreased stroke strength, another technique

to convey higher transparency is to use fewer lines by

transitioning from cross hatching to parallel hatching.

We implement this approach by using the alpha value

from the X-Toon texture as the probability for replac-

ing a point with a cross instead of a line. We multiply

that probability by a parameter

h ∈ [0,1]

, allowing us

to inﬂuence the proportion of cross-hatched to parallel-

hatched points. This parameter also makes it possible

to create results that exclusively use parallel hatching.

Curvature-controlled Line-length.

So far, we re-

place each point with equally long, straight lines. Put

together, they already create a good illustration of

curved surfaces because the lines are short enough

that they are a good approximation of the direction of

the surface below them. On areas of high curvature

on the other hand, the surface direction changes faster.

That means a line that is as long as the lines on low-

curvature areas is not a good surface approximation

anymore. With curvature information at each point,

we can align for that by decreasing the length of lines

in high-curvature areas (Fig. 5). In addition to line

length modiﬁcation through curvature, we also modify

the line length with a randomness factor to account for

the differences in line length when drawing by hand.

4 RESULTS & DISCUSSION

This section present and discusses various application

examples achieved by using our art-directable render-

ing technique.

Application Examples.

Fig. 7 shows an overview

of different point cloud stylization variants obtained

by our technique. The variety covers a range from

blueprint rendering, gray-scale charcoal and ballpoint

pens to classic pencil drawing or pointillism. Our

technique allows for modiﬁcation of all provided styl-

ization parameters and presents them interactively to

support the demands of artists.

Performance Considerations.

We tested the ren-

dering performance of a prototypical implementation

GRAPP 2022 - 17th International Conference on Computer Graphics Theory and Applications

224

(a) Naive Rendering (b) Back Face enabled (c) Depth Pass enabled (d) Difference of (b) and (c)

Figure 6: The Stanford dragon rendered as 3D point cloud (a) without back face culling, (b) with back face culling, and (c) with

an additional depth pass. Points—or strokes—that are hidden due to the depth pass are highlighted in red in (d).

using three point clouds of different complexity rang-

ing from approx. 3.5 million points (LiDAR, Statue)

over 5.7 million points to 11 million points (Angel

statue, sampled 3D mesh). The ﬁrst point cloud with

3.5 million points resulted in a frametime of about

30 milliseconds, with 5.7 million points we measured

60 milliseconds and for 11 million points we obtain

a frametime of 100 milliseconds. The performance

test was conducted using an AMD Radeon 5700XT

with 8 GB VRAM on a Ryzen 5 CPU with 3.6 GHz

and 32 GB RAM rendering at a viewport resolution

2947 × 1661

pixels. The run-time performance de-

pends mainly on the geometric complexity of the 3D

scene and decreases approx. linearly with the number

of points to process. However, while a high point den-

sity is useful for the depth pass to avoid gaps between

points it is not conducive to the overall visual quality

of the hatching effect, i.e., the generated strokes are

hardly distinguishable.

Discussion.

Due to the nature of point clouds, espe-

cially when created using consumer LiDAR scanners,

outliers from the depth pass (remember, all points are

used for this pass) sometimes cover strokes and might

not be desired. Similarly, the point cloud density is

typically rather irregular, which sometimes leads to

strokes bleeding from the background into the fore-

ground. Both effects can be mitigated by tweaking the

polygon offset as well as the adjusting the point size

within the depth pass. Furthermore, these effects are

not inherently wrong and are often also desirable for

less dense point clouds.

The same applies for strokes which leave an ob-

ject’s expected silhouette or contour (over-shooting of

strokes). For us, this was mostly visually pleasing and

also a welcome differentiator from image-based ren-

derings. There might be cases where this is not desired

or not ﬁtting a speciﬁc style though. Since point clouds

are never expected to be two manifold (i.e., watertight

or closed) and more often used for capturing open

environments than for solid and closed objects, we

render the mentioned drawbacks as non-problematic

exceptions of our technique. The distinctive styles

can be easily achieved with only minor conﬁgurations

within our art-directed parameterization and the tech-

nique can be implemented without exotic extension or

groundbreaking graphics APIs.

5 CONCLUSIONS

This paper presents a real-time non-photorealistic ren-

dering technique for art-directed hatches of 3D point

clouds. By extending the X-Toon approach of Barla

et al., it facilitates various conﬁgurations yielding dif-

ferent stylization results for potentially massive 3D

point clouds that are obtained by

LiDAR

acquisition

or sampling polygonal 3D meshes. The current state

provides basis to develop and apply more advanced

stylization techniques for 3D point clouds. For exam-

ple, to synthesize more sophisticated stylized render-

ings, our technique can be extended to support out-

lines (Rosenthal and Linsen, 2008) or even suggestive

contours (Proen

c¸

a et al., 2008) using the derived curva-

ture information (M

erigot et al., 2009; Lu et al., 2020).

Further, level-of-detail approaches for geometry gen-

eration stage can be implemented enabling seamless

level-of-abstraction transitions (Semmo and D

ollner,

2014). Furthermore, the depth pass could be improved

by using oriented point splats (Anjos et al., 2018).

ACKNOWLEDGMENTS

We thank the anonymous reviewers for their valu-

able feedback. This work was partially funded by the

German Federal Ministry of Education and Research

(

BMBF

) through grants 01IS18092 (“mdViPro”) and

01IS19006 (“KI-LAB-ITSE”).

A Non-Photorealistic Rendering Technique for Art-directed Hatching of 3D Point Clouds

225

(a) Cross Hatched

(b) Colorful Contour

(d) Two-Colored Dark

(e) White and Black Chalk

(f) Blueprint

(g) iPad Scan, Hatched

(h) iPad Scan, Cross Hatched

(i) iPad Scan, Crayons

(j) Statue (k) Temple

(l) Tree with Leaves

(m) Skull

(a) Cross Hatched

(b) Colorful Contour

(d) Two-Colored Dark

(e) White and Black Chalk

(f) Blueprint

(g) iPad Scan, Hatched

(h) iPad Scan, Cross Hatched

(i) iPad Scan, Crayons

(j) Statue (k) Temple

(l) Tree with Leaves

(m) Skull

(a) Cross Hatched

(b) Colorful Contour

(d) Two-Colored Dark

(e) White and Black Chalk

(f) Blueprint

(g) iPad Scan, Hatched

(h) iPad Scan, Cross Hatched

(i) iPad Scan, Crayons

(j) Statue (k) Temple

(l) Tree with Leaves

(m) Skull

(a) Cross Hatched

(b) Colorful Contour

(d) Two-Colored Dark

(e) White and Black Chalk

(f) Blueprint

(g) iPad Scan, Hatched

(h) iPad Scan, Cross Hatched

(i) iPad Scan, Crayons

(j) Statue (k) Temple

(l) Tree with Leaves

(m) Skull

(a) Cross Hatched

(b) Colorful Contour

(d) Two-Colored Dark

(e) White and Black Chalk

(f) Blueprint

(g) iPad Scan, Hatched

(h) iPad Scan, Cross Hatched

(i) iPad Scan, Crayons

(j) Statue (k) Temple

(l) Tree with Leaves

(m) Skull

(a) Cross Hatched

(b) Colorful Contour

(d) Two-Colored Dark

(e) White and Black Chalk

(f) Blueprint

(g) iPad Scan, Hatched

(h) iPad Scan, Cross Hatched

(i) iPad Scan, Crayons

(j) Statue (k) Temple

(l) Tree with Leaves

(m) Skull

(a) Cross Hatched

(b) Colorful Contour

(d) Two-Colored Dark

(e) White and Black Chalk

(f) Blueprint

(g) iPad Scan, Hatched

(h) iPad Scan, Cross Hatched

(i) iPad Scan, Crayons

(j) Statue (k) Temple

(l) Tree with Leaves

(m) Skull

(a) Cross Hatched

(b) Colorful Contour

(d) Two-Colored Dark

(e) White and Black Chalk

(f) Blueprint

(g) iPad Scan, Hatched

(h) iPad Scan, Cross Hatched

(i) iPad Scan, Crayons

(j) Statue (k) Temple

(l) Tree with Leaves

(m) Skull

(a) Cross Hatched

(b) Colorful Contour

(d) Two-Colored Dark

(e) White and Black Chalk

(f) Blueprint

(g) iPad Scan, Hatched

(h) iPad Scan, Cross Hatched

(i) iPad Scan, Crayons

(j) Statue (k) Temple

(l) Tree with Leaves

(m) Skull

(a) Cross Hatched

(b) Colorful Contour

(d) Two-Colored Dark

(e) White and Black Chalk

(f) Blueprint

(g) iPad Scan, Hatched

(h) iPad Scan, Cross Hatched

(i) iPad Scan, Crayons

(j) Statue (k) Temple

(l) Tree with Leaves

(m) Skull

(a) Cross Hatched

(b) Colorful Contour

(d) Two-Colored Dark

(e) White and Black Chalk

(f) Blueprint

(g) iPad Scan, Hatched

(h) iPad Scan, Cross Hatched

(i) iPad Scan, Crayons

(j) Statue (k) Temple

(l) Tree with Leaves

(m) Skull

Figure 7: Various results of our NPR rendering-technique. A point cloud scan of the Stanford bunny was used in (a)

to (f). The point clouds in (g) to (i) have are actual point clouds captured using an iPad. In (j) to (m) the points are

sampled from 3D polygonal meshes and rendered using our NPR technique. All results are captured from our C++ imple-

mentation, rendered interactively and in real-time (

60Hz+

in Full-HD and QHD) with a temporal coherence not possible

using image-based approaches. Meshes used for (j) and (k) are licensed under CC Attribution and have been created by

noe-3d.at (sketchfab.com/3d-models/grabﬁgur-cfec3a9d71db42978d4a377a0e8f1154) and Robin Butler (sketchfab.com/3d-

models/roman-temple-f5efe108fb19419ab985ef69d1762839) respectively.

GRAPP 2022 - 17th International Conference on Computer Graphics Theory and Applications

226

REFERENCES

Anjos, R. K., Ribeiro, C. S., Lopes, D. S. o., and Pereira,

J. a. M. (2018). Stroke-based splatting: An efﬁcient

multi-resolution point cloud visualization technique.

Vis. Comput., 34(10):1383––1397.

Ankermann, P., Swierad, O., and Krzyscin, K. (2021). The

tech and art of cyberspaces in cyberpunk 2077. In

ACM SIGGRAPH 2021 Talks, SIGGRAPH ’21, pages

21:1–2. Association for Computing Machinery.

Awano, N., Nishio, K., and Kobori, K.-i. (2010). Interactive

stipple rendering for point clouds.

Barla, P., Thollot, J., and Markosian, L. (2006). X-toon: An

extended toon shader. In Proc. ACM NPAR, NPAR ’06,

page 127–132. Association for Computing Machinery.

Cao, X., Shi, B., Okura, F., and Matsushita, Y. (2021).

Normal integration via inverse plane ﬁtting with min-

imum point-to-plane distance. In Proceedings of the

IEEE/CVF Conference on Computer Vision and Pat-

tern Recognition, pages 2382–2391.

Grabli, S., Turquin, E., Durand, F., and Sillion, F. X. (2004).

Programmable style for npr line drawing. In Keller, A.

and Jensen, H. W., editors, Eurographics Workshop on

Rendering. The Eurographics Association.

Kang, D., Chung, J.-M., Seo, S.-H., Choi, J.-S., and Yoon,

K.-H. (2009). Detail-adaptive toon shading using

saliency. In 2009 Second International Conference

in Visualisation, pages 16–20.

Kaplan, M., Gooch, B., and Cohen, E. (2000). Interactive

artistic rendering. In Proc. 1st International Sympo-

sium on Non-Photorealistic Animation and Rendering,

pages 67–74.

Kim, Y., Yu, J., Yu, X., and Lee, S. (2008). Line-art il-

lustration of dynamic and specular surfaces. In ACM

SIGGRAPH Asia 2008 Papers, SIGGRAPH Asia ’08,

pages 156:1–10. ACM.

Lu, D., Lu, X., Sun, Y., and Wang, J. (2020). Deep feature-

preserving normal estimation for point cloud ﬁltering.

CoRR, abs/2004.11563.

Markosian, L., Meier, B. J., Kowalski, M. A., Holden, L. S.,

Northrup, J. D., and Hughes, J. F. (2000). Art-based

rendering with continuous levels of detail. In Pro-

ceedings of the 1st International Symposium on Non-

Photorealistic Animation and Rendering, NPAR ’00,

pages 59—-66. Association for Computing Machinery.

erigot, Q., Ovsjanikov, M., and Guibas, L. (2009). Robust

voronoi-based curvature and feature estimation. In

2009 SIAMACM Joint Conference on Geometric and

Physical Modeling, SPM ’09, pages 1—-12. Associa-

tion for Computing Machinery.

Nienhaus, M. and D

ollner, J. (2003). Sketchy drawings:

A hardware-accelerated approach for real-time non-

photorealistic rendering. In ACM SIGGRAPH 2003

Sketches & Applications, SIGGRAPH ’03, page 1.

ACM.

Phong, B. T. (1975). Illumination for computer generated

pictures. Commun. ACM, 18(6):311—-317.

Praun, E., Hoppe, H., Webb, M., and Finkelstein, A. (2001).

Real-time hatching. In Proc. ACM SIGGRAPH, pages

579–584. ACM.

Proen

c¸

a, J., Jorge, J. A., and Sousa, M. C. (2008). Suggestive

contours over point-set implicits. In GRAPP, pages

171–180. INSTICC - Institute for Systems and Tech-

nologies of Information, Control and Communication.

Rosenthal, P. and Linsen, L. (2008). Image-space point

cloud rendering. In Proceedings of Computer Graphics

International (CGI) 2008, pages 136–143. Association

for Computing Machinery.

Runions, A., Samavati, F., and Prusinkiewicz, P. (2007).

Ribbons: A representation for point clouds. The Visual

Computer, 23(9-11):945–954.

Sabbadin, M., Palma, G., Banterle, F., Boubekeur, T., and

Cignoni, P. (2019). High dynamic range point clouds

for real-time relighting. Computer Graphics Forum,

38(7):513–525.

Schmid, J., Senn, M. S., Gross, M., and Sumner, R. W.

(2011). Overcoat: An implicit canvas for 3d painting.

ACM Trans. Graph., 30(4):28:1–10.

Sch

utz, M., Kerbl, B., and Wimmer, M. (2021). Rendering

point clouds with compute shaders and vertex order

optimization. Computer Graphics Forum, 40:115–126.

Sch

utz, M. and Wimmer, M. (2015). High-Quality Point

Based Rendering Using Fast Single Pass Interpolation.

In Guidi, G., Scopigno, R., and Brunet, P., editors,

International Congress on Digital Heritage - Theme 2

- Computer Graphics And Interaction. IEEE.

Semmo, A. and D

ollner, J. (2014). Image ﬁltering for inter-

active level-of-abstraction visualization of 3d scenes.

In Proceedings International Symposium on Computa-

tional Aesthetics in Graphics, Visualization, and Imag-

ing (CAe), pages 5–14.

Smith, A. R. (1984). Plants, fractals, and formal languages.

In Proc. ACM SIGGRAPH, SIGGRAPH ’84, pages

1–10. ACM.

Sousa, M. C. and Buchanan, J. W. (1999). Computer-

generated graphite pencil rendering of 3d polygonal

models. Computer Graphics Forum.

Webb, M., Praun, E., Finkelstein, A., and Hoppe, H. (2002).

Fine tone control in hardware hatching. In NPAR 2002:

Second International Symposium on Non Photorealis-

tic Rendering, pages 53–58.

Zakaria, N. and Seidel, H.-P. (2004). Interactive stylized

silhouette for point-sampled geometry. In Proceed-

ings of the 2nd International Conference on Computer

Graphics and Interactive Techniques in Australasia

and South East Asia, GRAPHITE ’04, pages 242—-

249, New York, NY, USA. Association for Computing

Machinery.

Zhang, L., Sun, Q., and He, Y. (2014). Splatting lines: An

efﬁcient method for illustrating 3d surfaces and vol-

umes. In Proceedings of the 18th Meeting of the ACM

SIGGRAPH Symposium on Interactive 3D Graphics

and Games, I3D ’14, pages 135—-142. Association

for Computing Machinery.

Zheng, M., Milliez, A., Gross, M., and Sumner, R. W.

(2017). Example-based brushes for coherent stylized

renderings. In Proceedings of the Symposium on Non-

Photorealistic Animation and Rendering, NPAR ’17,

pages 3:1–10. ACM.

A Non-Photorealistic Rendering Technique for Art-directed Hatching of 3D Point Clouds

227