Appealing Female Avatars from 3D Body Scans: Perceptual Effects of
Stylization
Reuben Fleming
1,2
, Betty J. Mohler
2
, Javier Romero
3
, Michael J. Black
3
and Martin Breidt
2
1
Sheffield Hallam University, Sheffield, U.K.
2
Max Planck Institute for Biological Cybernetics, T
¨
ubingen, Germany
3
Max Planck Institute for Intelligent Systems, T
¨
ubingen, Germany
Keywords:
3D Body Scan, Stylization, Avatar, Perception, Virtual Character.
Abstract:
Advances in 3D scanning technology allow us to create realistic virtual avatars from full body 3D scan data.
However, negative reactions to some realistic computer generated humans suggest that this approach might
not always provide the most appealing results. Using styles derived from existing popular character designs,
we present a novel automatic stylization technique for body shape and colour information based on a statistical
3D model of human bodies. We investigate whether such stylized body shapes result in increased perceived
appeal with two different experiments: One focuses on body shape alone, the other investigates the additional
role of surface colour and lighting. Our results consistently show that the most appealing avatar is a partially
stylized one. Importantly, avatars with high stylization or no stylization at all were rated to have the least
appeal. The inclusion of colour information and improvements to render quality had no significant effect on
the overall perceived appeal of the avatars, and we observe that the body shape primarily drives the change
in appeal ratings. For body scans with colour information, we found that a partially stylized avatar was most
effective, increasing average appeal ratings by approximately 34%.
1 INTRODUCTION
Virtual avatars are frequently used in games, vir-
tual worlds and for online communications. How an
avatar is perceived by others is considered extremely
important but creating a highly detailed and realistic
avatar does not necessarily produce appealing results
(Boberg et al., 2008). Appeal, one of the twelve prin-
ciples of animation (Johnston and Thomas, 1981) is
commonly used to describe well designed, interesting
and engaging characters. In the same way animated
movie and game characters require appeal in order
for people to engage with them, virtual avatars also
require appeal in order to engage others.
In this paper, we explore different ways in which
we can increase the appeal of 3D body scans via styl-
ization. In particular, we explore which styles make
the virtual avatars most appealing, and what is the op-
timal amount of stylization for the most successful
ones. We also examine the role of other factors in the
appeal of the final stylized avatar, namely the realism
of the renders and the appearance of the original sub-
ject. For this, we acquire 3D body scans of real people
using a state-of-the-art capturing system and automat-
ically create virtual avatars with different styles and
degrees of stylization. We render these avatars with
and without colour information and also improve the
render quality to simulate varying levels of realism.
We then study, in multiple experiments, which factors
affect the perceived appeal of these avatars.
2 BACKGROUND
Human-like virtual characters are becoming more and
more present in our lives. Since virtual environments
often try to replicate the real world, a natural design
choice for these virtual characters is to make them as
real as possible. This is becoming easier thanks to
the improvements of 3D capturing systems in terms of
speed, accuracy, quality and price. However, not all
virtual characters are designed with reality in mind.
Highly stylized avatars, such as the ones used by the
game consoles Xbox and Wii, seem to suggest that
people find the stylized look very appealing. The
stylized approach to avatar creation has traditionally
required the user to manually define and personalize
their avatar; creating a true resemblance of that person
Fleming, R., Mohler, B., Romero, J., Black, M. and Breidt, M.
Appealing Female Avatars from 3D Body Scans: Perceptual Effects of Stylization.
DOI: 10.5220/0005683903330343
In Proceedings of the 11th Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2016) - Volume 1: GRAPP, pages 335-345
ISBN: 978-989-758-175-5
Copyright
c
2016 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
335
is hard to achieve. Game studios are already attempt-
ing to cut out the manual work associated with this
process by capturing gamers’ features and applying
them to in-game characters. For example NBA 2K15
(Take-Two Interactive, 2014) attempts to capture the
player’s face while Kinect Sports Rivals (Microsoft,
2014) captures the entire body. Extracting informa-
tion about the player is currently achieved using pe-
ripherals such as Microsoft’s Kinect 2 and Sony’s PS4
camera.
Despite the availability of both high quality, real-
istic body scans and increasingly convincing avatars
from consumer peripherals, highly stylized characters
are still extremely popular and can be seen in many
animated movies, games and even web-based support
services. Animation companies such as Disney, Blue
Sky and Pixar tend to aim for a highly stylized look
when designing characters. Attempts to increase real-
ism in stylized characters have often led to negative
reactions from viewers (Melina, 2011; Chaminade
et al., 2007). Examples of this include films such as
Polar Express (Zemeckis, 2004), Mars Needs Moms
(Wells, 2011), and The Adventures of Tintin (Spiel-
berg, 2011).
Furthermore, negative reactions to the almost, but
not quite, photo-realistic characters seen in some film
and game productions (Chaminade et al., 2007) and
the human-like robots that inspired the “Uncanny Val-
ley” hypothesis (Mori et al., 2012) suggest that styl-
izing (therefore decreasing realism), rather than at-
tempting to improve realism, could actually produce
more appealing results. For example, (Inkpen and
Sedlins, 2011) conducted a survey into peoples’ com-
fort when communicating with avatars and found that
although respondents were comfortable interacting
with both realistic and cartoon looking avatars, some
avatars rated highly realistic were also felt to be eerie
or creepy. It is therefore possible that, even if one cre-
ated a photo-real digital replica of a human, a stylized
version could still be more appealing.
Experiments performed by (Seyama and Na-
gayama, 2007), where they morphed between artifi-
cial and real faces, suggest that stylization should be
done consistently in order to avoid viewer discom-
fort as abnormal features (e.g. cartoon eyes applied
to a human head) produced the lowest pleasantness
scores in their experiments. These conclusions match
similar findings by (MacDorman et al., 2009) who
found that a mismatch between the size and texture
of the eyes and face was especially prone to mak-
ing a character look eerie. Interestingly, again for
faces, results by (McDonnell et al., 2012) suggest that
both highly realistic and highly abstract render styles
were both considered appealing but render styles in-
between the two were considered unappealing. Re-
cently (Zell et al., 2015) found that while face shape
was important for portraying realism, it was the ma-
terial specifically the albedo map that played the
key role in influencing the perceived appeal of CG
characters and a stylized smooth looking skin texture
(obtained via blurring) was perceived as being more
appealing than a realistic skin texture.
To go beyond previous work on face appeal and
investigate the perceptual effects of stylization on
full-bodied avatars, we conducted two separate ex-
periments. The aim of the first experiment was to
investigate the effects of stylization when applied to
a realistic average female body shape containing no
colour information. This allowed us to focus purely
on body shape. In the second experiment we stylized
3D body scans of real people. The aim here was to
assess the effect of the subjects appearance (both in
terms of body shape and colour information) and ren-
der quality on perceived appeal and realism. Overall
the aim of these experiments is to better understand
how to create appealing female avatars from 3D body
scans.
3 STYLIZATION OF 3D BODY
SCANS
3.1 3D Body Scanning
To represent realistic human body shape and appear-
ance we first captured detailed 3D body scan and
colour information from 2 female actors (see Fig. 3)
in an A-pose using a full body 3D scanning system
(3dMD USA, 2014).
The system (see Fig. 1) is composed of 22 stereo
units, each including a pair of black and white cam-
eras observing a projected speckle light pattern, and
one 5 megapixel colour camera that captures the body
appearance 2 ms after the speckle image, synchro-
nized with white flash units. The system provides
very good coverage of the entire human body and can
resolve the 3D locations of a point on the body to ap-
proximately 1 mm.
Raw scans contain noise, occasional holes and are
difficult to manipulate. Consequently we registered
a common 3D body template to the scans in order
to obtain a coherent mesh topology for each. For
this we used a statistical 3D body model that com-
pactly parametrises body deformations automatically
in terms of shape changes due to subjects’ identity
and body pose (Hirshberg et al., 2012). The model is
based on approximately 2100 body scans of women
GRAPP 2016 - International Conference on Computer Graphics Theory and Applications
336
Figure 1: Taking a full body scan in the 3D scanning sys-
tem.
from the US and EU CAESAR dataset (Robinette
et al., 2002).
Through this process we obtained 3D geometry
that resembles the original 3D scans but also shares
a common layout. We call these meshes registrations.
Since registrations share mesh topology, their vertex
positions fully describe the body geometry. There-
fore, morphing between two or more bodies can be
performed as a weighted sum of the corresponding
vertex positions across registrations.
Once registered, we extracted both albedo and
pre-lit colour maps (Bogo et al., 2014) from each
scan (using the UV texture coordinate layout defined
in the template). The albedo maps were illuminated
using simple point lights within 3ds Max 2014 (Au-
todesk, USA). The pre-lit colour maps (Fig. 5 right),
provided a more realistic look and, unlike the albedo
maps (Fig. 5 left), required no CG lighting due to the
lighting information already being embedded within
the maps.
3.2 Shape Stylization
For our initial experiment, five female style tem-
plates were created in the digital sculpting applica-
tion Zbrush (Pixologic, USA) (Fig. 2). The two most
appealing styles were then used again for our second
experiment.
The style templates were created by a trained 3D
artist and were modelled to match the body propor-
tions of existing stylized designs (see Tab. 1). The
style templates used the same topology as the regis-
trations, therefore direct morphing between registra-
tions and style templates was made possible by sim-
ply interpolating vertex positions. While creating the
styles is a time-consuming manual task that requires
a skilled artist, it is worth noting that thanks to the
shared topology, this only needs to be done once and
can then be reused for all future scans.
Table 1: Female style template references.
Style Name Description
Marvel Typical female Marvel superheroine
(Marvel Entertainment, USA)
Disney “Rapunzel” from the animated
movie “Tangled” (Disney, 2010)
Sony “Sam Sparks” from the animated
movie “Cloudy with a Chance of
Meatballs” (Sony, 2009)
Pixar “Princess Merida” from the ani-
mated movie “Brave” (Pixar, 2012)
Barbie “Barbie” doll (Mattel, USA)
Figure 2: Style templates created from character reference.
3.3 Body Shape Generation
For experiment 1 we used an average female body
shape M (Fig. 3 top left). This shape was computed
by averaging the n 2100 original registrations O
i
(as
described in section 3.1): M =
1
n
n
i=1
O
i
.
In experiment 2, rather than stylize the average fe-
male body shape, we stylized registrations F obtained
from 3D body scans of real people. In order to main-
tain the characteristic body features (even when fully
stylized) we augmented the resulting morph between
registration F
i
and template S
j
with data derived from
the registration: Half of the Euclidean difference be-
tween the mean body shape M and the registration F
i
was added back to the current body style template S
j
(right term of Eq. 1). See Fig. 3 for an illustration of
the process.
We obtain the final stylized avatar shape N
i, j|α
used for one trial (with identity i, style j and styliza-
tion level α [0. . . 1]) by linear interpolation between
the original registration F
i
and the current style tem-
plate S
j
, augmented by the individual body character-
istics:
N
i, j|α
= (1 α) F
i
+ α
S
j
+
1
2
(F
i
M)
(1)
This helped to maintain the more subtle characteris-
tics of the original body shape while keeping the over-
all proportions of the stylized models.
Appealing Female Avatars from 3D Body Scans: Perceptual Effects of Stylization
337
+
+
+
-1 1
0.5
1
α
(1 α )
M
S
S
2
F
1
F
2
N
2,2
| α=1
1,2
| α=1
N
Figure 3: Stylization process for style S
2
(Disney).
Eq. 1 was implemented using the Morpher modi-
fier in 3ds Max, morphing the scans from their origi-
nal state to the 100% stylized model (see Fig. 3 bot-
tom for examples). Two intermediate states were also
generated at 33% and 66% stylization.
3.4 Stylization of Albedo and Pre-lit
Colour Maps
For experiment 2, we used the albedo and pre-lit
colour maps extracted from our scans. These required
some manual clean-up work as they contained vi-
sual marker information (recorded for a different pur-
pose) and had some missing data (Fig. 4). This was
achieved using basic retouching tools and techniques
in Photoshop (Adobe, USA). Note that normally this
step would not be necessary as subjects are scanned
without markers.
Figure 4: Original albedo map (left), cleaned up and grey
body suit added (centre), stylized result (right).
As all actors wore swimwear during scanning, a
grey body suit template was placed over each map
(Fig. 4 centre) to avoid participants becoming dis-
tracted by the clothing. Due to the shared topol-
ogy, and thus the shared UV texture layout, this only
needed to be done once for all characters. To match
the realism of the pre-lit colour maps, a realistic ver-
sion of the suit was rendered using area lights and
global illumination in the V-Ray 2.0 renderer (Chaos
Group, Bulgaria).
The albedo and pre-lit colour maps were also styl-
ized in order to recreate a similar look to most of
the style references. A typical feature film anima-
tion look is often represented through heavily satu-
rated colours and soft surface details. To achieve
this, our colour stylization technique was imple-
mented as a Photoshop action and involved Photo-
shop’s edge-preserving Surface Blur filter and in-
creasing the colour Vibrancy (increasing the satura-
tion of less saturated colours), controlled by masking
regions (Fig. 4 right).
Both the albedo and pre-lit colour maps had a res-
olution of 1024 × 1024 pixels. Parameter α in Eq. 1
was used to linearly blend between the normal and
stylized versions of these maps.
Figure 5: Avatar using CG lit albedo map (left), and pre-lit
colour map with embedded lighting (right).
4 EXPERIMENT 1
We ran a 15 minute experiment with 18 Participants
(10 female, average age = 26.5 yr, SD = 7.5). The ex-
periment was conducted in accordance with the Dec-
laration of Helsinki. All participants signed informed
consent and were financially compensated.
4.1 Experimental Design
For our first experiment we asked participants for
their judgements of appeal on a 7-pt Likert scale when
observing an average female body shape, as well as
that shape being morphed into 5 different style tem-
plates (Fig. 6) at 3 levels of stylization (33, 66, 100%).
This resulted in a total of 16 avatars (15 for the styl-
ized body and 1 for the average body shape). Stimuli
were presented in a random order. The avatar featured
no colour information and was rendered in real-time
within the Unity 4 game engine (Unity Technologies,
Denmark).
GRAPP 2016 - International Conference on Computer Graphics Theory and Applications
338
Figure 6: Styles used in Experiment 1.
Participants were instructed to answer quickly and
simply provide their first impression. Participants
were also given an in-depth definition of appeal be-
fore they began:
High appeal: “you find the person extremely appeal-
ing. This may include finding the person ex-
tremely engaging, likeable and/or generally pleas-
ing to the eye.
Average appeal: “you neither find the person appeal-
ing nor unappealing. They are acceptable but
essentially you are indifferent, finding them nei-
ther engaging nor disengaging, likeable nor un-
likeable.
Low appeal: “you find the person extremely unap-
pealing. This may include finding the person ex-
tremely disengaging, unlikeable and/or repellent.
4.2 Experiment 1 Results
A partially stylized avatar was often rated as more ap-
pealing than the average female (where the amount
of stylization for optimal appeal depends on the style,
see Fig. 7 and Tab. 2).
We ran a repeated measures ANOVA on the rat-
ing of appeal of avatars in a within subject factors de-
Table 2: Mean appeal ratings by stylization level and style
(standard deviations in parentheses).
Style 33% stylized 66% stylized 100% stylized
Marvel 4.94 (1.211) 5.00 (1.085) 4.17 (1.724)
Disney 4.72 (1.127) 4.83 (1.425) 3.67 (1.847)
Barbie 4.39 (1.195) 4.00 (1.572) 2.33 (1.138)
Pixar 4.39 (1.092) 3.44 (1.504) 2.17 (1.043)
Sony 2.22 (1.003) 1.56 (0.784) 1.28 (0.461)
Percentage Stylized
original 33% 66% 100%
Average Appeal Rating (1-7 Likert)
1
2
3
4
5
6
Average Female
Marvel
Disney
Barbie
Pixar
Sony
Figure 7: Appeal ratings on a 7-pt Likert scale by style
and stylization level (n=18). Ratings for the average female
have been included in this graph as a reference. Error bars
represent one standard error of the mean in all figures.
sign with two factors: Style (5 levels: Marvel, Disney,
Sony, Pixar, Barbie) and Percent Stylized (3 levels:
33, 66, 100%).
There was a significant effect of Percent Stylized
on appeal ratings, F(2,34) = 18.33, p < 0.001, η
2
p
= 0.519. 100% stylized had significantly lower ap-
peal ratings than both 33% and 66% while both 33%
and 66% stylized avatars were found to be more ap-
pealing than the original body scan (as revealed by
post-hoc comparisons using Bonferroni adjustments,
(p < 0.001)).
There was also a significant effect of Style on ap-
peal ratings, F(4,68) = 63.03, p < 0.001, η
2
p
= 0.788.
The five styles did not have the same effect on ap-
peal ratings across percent stylized (as revealed by
post-hoc comparisons using Bonferroni adjustments).
Most notably the Sony style had a significantly lower
rating of appeal than all of the styles (p < 0.001) and
was always rated as less appealing than the original
body scan. The appeal ratings of the Marvel and
Disney styles did not significantly differ from each
other (p = 1.00) and appeal ratings of the Barbie and
Pixar styles also did not significantly differ from each
other (p = 0.907). All other comparisons between
styles resulted in significantly different appeal ratings
(p < 0.05).
These main effects were conditioned upon a sig-
nificant interaction between style and percent styl-
ized, F(8,136) = 3.02, p = 0.004, η
2
p
= 0.151. We
therefore tested for an effect of percent stylized for
each individual style. The effects of percent stylized
were significant for all styles (p < 0.05) except for the
Marvel style (p = 0.12). Not surprisingly, there was a
simple effect of style found when testing at each per-
cent stylized, driven by the low ratings of appeal for
Sony (p < 0.05). In addition, at 100% stylized the
Marvel and Disney appeal ratings were significantly
greater than both Barbie and Pixar appeal ratings (as
Appealing Female Avatars from 3D Body Scans: Perceptual Effects of Stylization
339
Figure 8: Styles featuring albedo maps and CG lighting.
revealed by post-hoc comparisons using Bonferroni
adjustments, (ps < 0.05)).
These results show that for all styles except Sony,
a partially stylized avatar leads to the greatest appeal
ratings and depending on that style either 33% or 66%
leads to optimal appeal. Excluding the Sony style,
an improvement upon the original appeal ratings for
the average female of approximately 29% could be
achieved by stylizing the body shape.
5 EXPERIMENT 2
Both the Marvel and Disney styles provided the most
significant improvements for appeal ratings in exper-
iment 1 and were therefore used to stylize the body
scans of real people in our second experiment.
5.1 Visual Stimuli
For experiment 2, 14 images were rendered in 3ds
Max at 500 × 1000 pixels using our CG lit albedo
maps, 7 rendered images per character: three percent-
ages at two styles each, plus the original body scans
(Fig. 8). Another 14 images were rendered in 3ds
Max using our pre-lit colour maps at the same res-
olution (Fig. 9).
5.2 Experimental Design
There were 61 Participants (35 female) with an aver-
age age of 28.08 yr (SD = 7.6) and none of the par-
ticipants had been involved in experiment 1. The ex-
Figure 9: Styles featuring pre-lit colour maps.
periment was conducted in accordance with the Dec-
laration of Helsinki. All participants signed informed
consent and were compensated. There was no time
limit assigned to the sessions but participants were in-
structed to answer quickly and provide their first im-
pression. The experiment lasted approximately one
hour.
There were two tasks in this experiment. First,
subjects provided appeal ratings as in Experiment 1.
Second, participants were asked were asked to dy-
namically create the most appealing and repelling
characters.
5.2.1 Task 1: Likert Scale Ratings
Experiment 2 was a repeated measures design and in-
volved rating avatars on a 5-pt Likert scale and had
three within subject factors: Actor Identity (2 levels:
F
1
, F
2
), Style (2 levels: Marvel, Disney) and Percent
Stylized (3 levels: 33, 66, 100%) and one between
subject factor of Render Quality (2 levels: albedo
maps with CG lighting (42 participants), pre-lit colour
maps without CG lighting (19 participants)).
In addition to appeal, we asked our participants six
further questions (Tab. 3) to see whether our styliza-
tion technique had a significant effect on other char-
acter traits
1
.
This resulted in a total of 7 questions being asked
for 14 avatars (2 actors × 2 styles × 3 stylization
1
Our questions were inspired by existing research, e.g.
appeal (McDonnell et al., 2012; McDonnell, 2012) and
sympathy (Hodgins et al., 2010), neuroticism inspired by
the Big Five personality traits (Johnson et al., 1993).
GRAPP 2016 - International Conference on Computer Graphics Theory and Applications
340
levels) + 2 original body scans. We additionally
asked the participants who saw the avatars with pre-
lit colour maps to rate realism. The avatars were pre-
sented in a random order. Questions were asked for
each avatar in the same order as in Table 3.
Table 3: Avatar questionnaire.
1 Do you feel as if you can sympathise with this per-
son?
2 Are you repelled by this person?
3 Would you accept help from this person?
4 Do you find this person engaging?
5 How appealing do you find this person?
6 How charming do you find this person?
7 Does this person come across as being neurotic?
5.2.2 Task 2: Interactive Creation of Appealing
Characters
In a second task, 37 participants were asked to specify
the exact level of stylization they found to be most
appealing/repelling by interactively adjusting a slider
in real-time within the Unity 4 game engine.
In addition to the stylized result controlled by the
slider, the participants were also shown the original
3D body scan next to the avatar they were manipu-
lating, giving them a non-stylized reference. Albedo
maps and CG lighting were used for this task. The
dependent variable was the chosen percentage of styl-
ization and the fixed factor was the style (Marvel and
Disney) and the actor identity (F
1
and F
2
). Each trial
always began with the slider value set to 0% (i.e.
showing the original body shape). This resulted in
four trials (two styles for each of the two actor iden-
tities). They were presented in random order. The
instructions were as follows:
“Make this character as appealing as possible.”,
and “Make this character as repelling as possible.
For this task participants were instructed to take
their time and be as accurate as possible. Participants
were also instructed to adjust the slider once more af-
ter deciding on a level of stylization to verify their
decision before submitting a rating.
5.3 Experiment 2 Results
For the most part responses to our questions show a
similar pattern across percent stylized: an increase in
rating at 33% and a decrease in rating at 100% (as
can be seen in Fig. 10). We decided to analyze only
appeal and realism ratings because appeal seems to
be representative of the majority of the questions in
Table 3. Indeed the mean Appeal rating did not sig-
nificantly differ from all other questions (p < 0.05)
(with the exceptions only being neurotic and repelling
which were inverted in scale). Realism appears to be
the only question that demonstrates a unique pattern
across percent stylized so we analyzed separately re-
alism ratings in the following section.
Percentage Stylized
original 33% 66% 100%
Average Rating (1-5 Likert)
1,5
2,0
2,5
3,0
3,5
4,0
sympathize with
repelled by
accept help
engaging
appealing
charming
neurotic
realistic
Figure 10: Ratings on a 5-pt Likert scale by stylization
level for all questions, collapsed across all styles, identi-
ties and render quality (n=61, n=19 for realism). Note: Re-
pelling and Neurotic ask for negative traits and therefore
have inverted values compared to all other questions asking
for positive traits.
5.3.1 Appeal Ratings Results
Average appeal rating for the original body scans was
2.69. The average appeal ratings for style templates
was 3.49 for Marvel and 3.19 for Disney (see Fig. 11
and Tab. 4).
Percentage Stylized
original 33% 66% 100%
Average Appeal Rating (1-5 Likert)
2.0
2.5
3.0
3.5
4.0
4.5
Disney
Marvel
Disney
Marvel
F
1
F
1
F
2
F
2
F
1
F
2
Figure 11: Appeal ratings on a 5-pt Likert Scale by style,
actor identity and stylization level, collapsed across both
render styles (n=61).
We ran a repeated measures ANOVA on the rat-
ing of appeal of characters with three within subject
factors: Actor Identity (2 levels, F
1
, F
2
), Style (2 lev-
els: Marvel, Disney) and Percent Stylized (3 levels:
33, 66, 100%) and one between subject factor of ren-
der quality (2 levels: albedo maps with CG lighting,
pre-lit colour maps without CG lighting).
Appealing Female Avatars from 3D Body Scans: Perceptual Effects of Stylization
341
Table 4: Appeal ratings per actor, style and stylization level
(standard deviations in parentheses).
Actor F
1
Actor F
2
Level Marvel Disney Marvel Disney
original 2.56 (1.13) 2.56 (1.13) 2.82 (1.02) 2.82 (1.02)
33% 3.52 (0.96) 3.36 (1.11) 3.62 (0.90) 3.49 (0.92)
66% 3.57 (1.02) 2.89 (1.13) 3.87 (1.02) 3.64 (1.05)
100% 2.93 (0.89) 2.62 (1.11) 3.34 (1.14) 2.89 (1.02)
Surprisingly, we found no significant effect of ren-
der style on appeal ratings (albedo maps with CG
lighting versus pre-lit colour maps without CG light-
ing stimuli), F(1,59) = 1.03, p = 0.315, η
2
p
= 0.017.
Similar to Experiment 1, we found a significant
effect of percent stylized on appeal ratings, F(2,118)
= 19.63, p < 0.001, η
2
p
= 0.250. In relative terms, our
method achieves approximately 34% improvement in
appeal ratings for the original body scans by stylizing
at 33% or 66% towards Marvel or Disney styles.
33% and 66% do not significantly differ from each
other on appeal ratings (p = 1.0), while 100% styl-
ized is significantly lower rated on appeal than both
33% and 66% (p < 0.001, as revealed through post-
hoc comparisons using Bonferroni adjustments)
We also found a significant effect of style on ap-
peal ratings, F(1,59) = 16.24, p < 0.001, η
2
p
= 0.216.
Marvel is significantly higher rated on appeal than
Disney (as revealed through post-hoc comparisons us-
ing Bonferroni adjustments, p < 0.001). Finally, we
found a significant effect of actor identity, F(1,59)=
8.66, p < 0.005, η
2
p
= 0.128. A stylized female F
1
is significantly lower rated on appeal than a stylized
female F
2
(as revealed through post-hoc comparisons
using Bonferroni adjustments, p < 0.001).
These effects suggest that all factors (style, actor
identity and percent stylized) influence the ratings of
appeal. These main effects were conditioned upon a
significant interaction between Actor Identity, Style
and Percent Stylized, F(2,118) = 3.45, p = 0.035, η
2
p
= 0.055 and a significant interaction between Actor
Identity and Percent Stylized F(2,118) = 5.34, p <
0.01,η
2
p
= 0.083. The Disney F
1
at 66% seems to drive
the interaction between Actor Identity, Style and Per-
cent Stylized. This one data point deviates from the
typical pattern of improved appeal at 33%, greater im-
provement at 66% and then a decrease in appeal at
100% stylized, as revealed through post-hoc compar-
isons using Bonferroni adjustments, p < 0.001.
5.3.2 Realism Ratings Results
As a reminder, we only asked realism ratings for the
avatars with pre-lit colour maps. Realism ratings for
the original body scans was on average 3.61. Aver-
age realism for the stylized avatar at 33% was 3.91,
at 66% was 3.54, and at 100% was 2.72 (see Fig. 12
and Tab. 5). Interestingly at 33% stylized there was
no decrease in ratings of realism as compared to the
original scan, at 66% only the Female F
1
Disney style
received a decreased realism rating and for 100% al-
most all stylized body scans received a significantly
lower realism rating.
We ran a repeated measures ANOVA on the rating
of realism of avatars with three within subject factors:
Actor Identity (2 levels, F
1
, F
2
), Style (2 Levels: Mar-
vel, Disney) and Percent Stylized (3 levels: 33, 66,
100). Since we had not asked for realism ratings with
the albedo colour map with CG lighting there is no
between subject factor for ratings of realism.
Percentage Stylized
original 33% 66% 100%
Average Realism Rating (1-5 Likert)
2.0
2.5
3.0
3.5
4.0
4.5
Disney
Marvel
Disney
Marvel
F
1
F
1
F
2
F
2
F
1
F
2
Figure 12: Realism ratings on a 5-pt Likert scale by style,
actor identity and stylization level (n=19).
Table 5: Realism ratings per actor identity, style and styl-
ization level (standard deviations in parentheses).
Actor F
1
Actor F
2
Level Marvel Disney Marvel Disney
original 3.47 (1.07) 3.47 (1.07) 3.74 (1.09) 3.74 (1.09)
33% 3.79 (0.79) 4.05 (0.71) 3.84 (0.96) 3.95 (0.78)
66% 3.63 (1.07) 2.95 (0.97) 4.05 (0.71) 3.53 (0.70)
100% 2.74 (0.73) 2.16 (0.50) 3.37 (0.68) 2.63 (0.83)
We found a significant effect of Percent Stylized,
F(2,36) =35.042, p < 0.001, η
2
p
= 0.661. There was a
significant difference between 33, 66 and 100% styl-
ized in realism (as revealed through post-hoc compar-
isons using Bonferroni adjustments, p < 0.05). This
was seen as a decline in realism ratings as stylization
percentage increased.
We found a significant effect of style, F(1,18) =
18.841 , p < 0.000, η
2
p
= 0.511. Marvel is rated to
have a significantly higher realism rating compared
to Disney (as revealed through post-hoc comparisons
using Bonferroni adjustments, p < 0.001). Similarly,
we also found a significant effect of actor identity,
F(1,18) = 29.439 ,p < 0.001, η
2
p
= 0.621. Consis-
GRAPP 2016 - International Conference on Computer Graphics Theory and Applications
342
tent with appeal ratings, a stylized female F
1
is rated
significantly lower for realism than a stylized female
F
2
(as revealed through post-hoc comparisons using
Bonferroni adjustments, p < 0.001).
These main effects were conditioned upon a sig-
nificant interaction between style and percent stylized
F(2,36) = 11.378,p < 0.001, η
2
p
= 0.387 and a signif-
icant interaction between actor identity and percent
stylized F(2, 36) = 6.610, p < 0.005, η
2
p
= 0.269. This
can be summarized by the styles not changing realism
ratings at 33% stylized, but as stylization increases the
styles and the actors are being perceived with a variety
of different ratings of realism. High levels of Marvel
stylization result in decreased realism ratings for F
1
but not so for F
2
, where the Marvel stylization seems
to have the same influence on realism ratings across
the actors. These results are consistent with our ap-
peal ratings and may shed some light on why 100%
stylization is seen as less appealing, because it is also
seen as less realistic.
5.3.3 Interactive Creation of Characters Results
When asked to interactively create the most appealing
character using sliders it was extremely rare that par-
ticipants chose either the original body scan (3%) or
the fully stylized avatar (5% of the trials). The aver-
age stylization amount chosen was 40.9% across ac-
tors and styles for an appealing character (see Fig. 13
and Tab. 6).
Due to this amodal distribution for the repelling
Figure 13: Original body scan of female F
1
(left), 38% Dis-
ney (center), 51% Marvel (right). Real-time images cap-
tured from within the Unity 4 game engine.
Table 6: Average stylization level in the method of adjust-
ment task for most appealing character per actor and style
(standard deviations in parentheses).
Style F
1
F
2
Marvel 50.77% (26.3) 45.56% (24.90)
Disney 37.65% (23.3) 29.71% (22.02)
characters we decided to present the histograms for
the percent stylized chosen (see Fig. 14). When
participants were asked to create the most repelling
avatar they chose the 100% stylized version for 58%
of the total trials and the original scan for 24% of the
total trials. This further confirms that the most appeal-
ing avatar is a partially, and specifically not a fully,
stylized one. These results are also consistent with
the repelling ratings from task 1. However, the de-
gree to which the original body scans and 100% styl-
ized body scans were rated as repelling was only just
above 2 on the 5-pt Likert scale (see Fig. 10). Im-
portant therefore to note is that original body scans
and 100% stylized avatars were not seen as extremely
repelling, but rather more repelling as compared to
partially stylized avatars.
Percentage Stylized
1009080706050403020100
Frequency
12.5%
10.0%
7.5%
5.0%
2.5%
0.0%
Percentage Stylized
1009080706050403020100
Frequency
60.0%
50.0%
40.0%
30.0%
20.0%
10.0%
0.0%
Figure 14: Frequency of stylization level chosen interac-
tively for maximum appealing (top) and repelling characters
(bottom), collapsed across actor identity and style (n=37).
6 DISCUSSION
In general, appeal ratings for the average female body
shape were highest when some level of stylization
was applied (usually 33% but in some cases 66%).
However, not all styles affect the resulting appeal
score in a similar manner. In experiment 1, the Dis-
ney and Marvel stylizations were found to be consis-
tently more appealing than the average female body
shape. Styles with exaggerated body shapes tended
to be rated lower than styles with more human-like
body shapes, with the Barbie and Pixar styles produc-
ing lower appeal ratings than the Disney and Marvel
styles. The Sony style (featuring the most exagger-
Appealing Female Avatars from 3D Body Scans: Perceptual Effects of Stylization
343
ated body shape) was rated as being consistently less
appealing than the average female body shape, re-
gardless of the level of stylization applied. This seems
to suggest a preference for human likeness when ob-
serving avatars and a dislike for extremely exagger-
ated proportions. In our second experiment involving
the stylization of body scans with colour information,
we again found that moderate levels of stylization are
effective at increasing appeal since 33% and 66% lev-
els both produced higher appeal ratings than the orig-
inal body scans. However, 100% stylized avatars in
most cases were rated no differently than the origi-
nal body scans. We also found that efforts to make
the avatars appear more realistic (through improve-
ments in render quality) had no significant impact on
appeal ratings. Considering our results, it could be
that in contradiction to (Zell et al., 2015), body shape
is playing a larger role in the influence of perceived
appeal than the inclusion (and modification) of sur-
face colour, however further testing is needed to bet-
ter understand this. Realism ratings were consistent
with appeal ratings and may provide some insight as
to why higher levels of stylization are less desirable
than partial stylization, as the highest level of styl-
ization was rated lowest in realism. Interesting, 33%
stylized body scans were not seen as less realistic as
compared to the original body scans. This could be
due to the stylization of the texture drawing attention
away from imperfections in the surface colour. The
fact that there is not an immediate decrease in realism
with stylization also is a result that further supports
the use of stylization to create more appealing avatars.
Additionally, a partially stylized avatar may not suffer
from the Uncanny Valley effect. We hope to further
investigate the perception of realism in more depth to
better understand the role realism plays in the appeal
of our avatars and specifically further investigate why
our original body scans receive similar realism ratings
as the 33% stylized avatars. A limitation of our find-
ings is the small number of templates and body reg-
istrations presented in this paper. We hope to extend
this work by investigating a larger variance in body
shape of actors (i.e. male participants, a wider range
of BMI and different age groups). Most importantly
we would like to increase the number of style tem-
plates that we use to stylize our avatars to be suitable
for a greater variance in the shape of humans.
7 SUMMARY
The stylization technique presented in this paper al-
lows for the semi-automatic creation of more ap-
pealing personalized avatars from 3D body scans for
use in games, virtual environments and online com-
munications.
For stylizing 3D body scans, the main steps are the
automatic registration of a statistical 3D body model
to the body scan data, followed by the stylization
through feature-preserving morphing with colour fil-
tering and blending. Our technique can be automat-
ically applied to any human body scan after registra-
tion of the scan to our 3D body template. The only
manual process involved is the sculpting of the style
templates. However, once created, these can be ap-
plied to an infinite number of body scans and based on
the individual templates, can be applied at the optimal
level of stylization required to maximize the appeal of
these body scans. The rest of the process is fully au-
tomated. Based on the results presented in this paper,
the proposed system can increase appeal ratings by
approximately 34% depending on the actor.
We conducted multiple experiments in which par-
ticipants were asked to rate the appeal of different
avatars. Our results demonstrate that some stylization
(approximately 37%) is perceived as most appealing
on average for our female actors. This suggests that
3D body scans can be made to look more appealing
with some type/level of stylization. Finally, when we
asked our participants to interactively create the most
appealing and repelling avatars from 3D body scans,
we found again that a partially stylized avatar is the
most appealing at approximately 41%. In contrast,
when considering the percent of stylization people
used to create repelling avatars, we found that 82% of
the chosen avatars were made to be either 100% styl-
ized or the original body scans themselves and that a
highly stylized avatar was most often chosen as the
most repelling avatar.
Our results show that partially stylized female 3D
body scans are perceived as most appealing as com-
pared to the original and fully stylized body scans.
ACKNOWLEDGEMENTS
We would like to thank Chris Ferguson, Joe Small-
wood, Anna Wellerdiek and Michael Geuss as well as
all the participants of our experiments and the anony-
mous reviewers.
GRAPP 2016 - International Conference on Computer Graphics Theory and Applications
344
REFERENCES
3dMD USA (2014). 3dMDbody System.
http://www.3dmd.com/3dMDbody/. Accessed:
2014-07-05.
Boberg, M., Piippo, P., and Ollila, E. (2008). Designing
avatars. In Proceedings of the 3rd International Con-
ference on Digital Interactive Media in Entertainment
and Arts, DIMEA ’08, pages 232–239, New York,
USA. ACM.
Bogo, F., Romero, J., Loper, M., and Black, M. J. (2014).
FAUST: Dataset and evaluation for 3D mesh regis-
tration. In Proceedings IEEE Conf. on Computer
Vision and Pattern Recognition (CVPR), Piscataway,
NJ, USA. IEEE.
Chaminade, T., Hodgins, J., and Kawato, M. (2007). An-
thropomorphism influences perception of computer-
animated characters actions. Social Cognitive and Af-
fective Neuroscience, 2(3):206–216.
Hirshberg, D., Loper, M., Rachlin, E., and Black, M.
(2012). Coregistration: Simultaneous alignment and
modeling of articulated 3d shape. In Fitzgibbon, A.,
Lazebnik, S., Perona, P., Sato, Y., and Schmid, C., ed-
itors, Computer Vision – ECCV 2012, volume 7577 of
Lecture Notes in Computer Science, pages 242–255.
Springer Berlin Heidelberg.
Hodgins, J., Jrg, S., O’Sullivan, C., Park, S. I., and
Mahler, M. (2010). The saliency of anomalies in ani-
mated human characters. ACM Trans. Appl. Percept.,
7(4):22:122:14.
Inkpen, K. M. and Sedlins, M. (2011). Me and my
avatar: Exploring users’ comfort with avatars for
workplace communication. In Proceedings of the
ACM 2011 Conference on Computer Supported Coop-
erative Work, CSCW ’11, pages 383–386, New York,
USA. ACM.
Johnson, J. A., Ostendorf, F., Johnson, J. A., Psycholog, D.,
and State, P. (1993). Clarification of the ve-factor
model with the abridged big five dimensional circum-
plex. Journal of Personality and Social Psychology,
pages 563–576.
Johnston, O. and Thomas, F. (1981). Disney Animation:
The Illusion of Life. Abbeville Press.
MacDorman, K. F., Green, R. D., Ho, C.-C., and Koch, C. T.
(2009). Too real for comfort? Uncanny responses to
computer generated faces. Computers in Human Be-
havior, 25(3):695–710.
McDonnell, R. (2012). Appealing virtual humans. In Kall-
mann, M. and Bekris, K., editors, Motion in Games,
volume 7660 of Lecture Notes in Computer Science,
pages 102–111. Springer Berlin Heidelberg.
McDonnell, R., Breidt, M., and B
¨
ulthoff, H. H. (2012).
Render me real: Investigating the effect of render style
on the perception of animated virtual humans. ACM
Trans. Graph., 31(4):91:1–91:11.
Melina, R. (2011). Is ‘Mars Needs Moms’ Too Real-
istic? http://www.livescience.com/33198-is-mars-
needs-moms-too-realistic.html. Accessed: 2014-06-
18.
Microsoft (2014). Kinect Sports Rivals. Computer Game.
Mori, M., MacDorman, K., and Kageki, N. (2012). The
Uncanny Valley [from the field]. IEEE Robotics Au-
tomation Magazine, 19(2):98–100.
Robinette, K. M., Blackwell, S., Daanen, H., Boehmer, M.,
and Fleming, S. (2002). Civilian american and euro-
pean surface anthropometry resource (CAESAR), fi-
nal report. volume 1. summary. Technical report, US
Air Force.
Seyama, J. and Nagayama, R. S. (2007). The uncanny val-
ley: Effect of realism on the impression of artificial
human faces. Presence: Teleoperators and Virtual En-
vironments, 16(4):337–351.
Spielberg, S. (2011). The Adventures of Tintin. Animated
Feature Film.
Take-Two Interactive (2014). NBA 2K15. Computer Game.
Wells, S. (2011). Mars Needs Moms. Animated Feature
Film.
Zell, E., Aliaga, C., Jarabo, A., Zibrek, K., Gutierrez, D.,
McDonnell, R., and Botsch, M. (2015). To stylize or
not to stylize?: The effect of shape and material styl-
ization on the perception of computer-generated faces.
ACM Trans. Graph., 34(6):184:1–184:12.
Zemeckis, R. (2004). The Polar Express. Animated Feature
Film.
Appealing Female Avatars from 3D Body Scans: Perceptual Effects of Stylization
345