Automated Facial Mark Creating Systems Replace Classical Geometric Morphometrics: An Example of How New Technology Can and Should Drive Avatar Creation in a Game Development Pipeline

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F44555601%3A13510%2F24%3A43898974" target="_blank" >RIV/44555601:13510/24:43898974 - isvavai.cz</a>

Result on the web

<a href="https://link.springer.com/chapter/10.1007/978-3-031-76812-5_26#citeas" target="_blank" >https://link.springer.com/chapter/10.1007/978-3-031-76812-5_26#citeas</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/978-3-031-76812-5_26" target="_blank" >10.1007/978-3-031-76812-5_26</a>

Result language

angličtina

Original language name

Automated Facial Mark Creating Systems Replace Classical Geometric Morphometrics: An Example of How New Technology Can and Should Drive Avatar Creation in a Game Development Pipeline

Original language description

In recent years, the field of artificial intelligence (AI) has witnessed significant advancements, particularly with the development of transformer architectures and pre-learned feature extraction techniques. Before this, thousands of high-quality scientific papers using biological landmarks (manually placed by a researcher) on faces have identified relations between facial morphology and various human ratings of them. Geometric Morphometrics provides a framework for quantifying and analyzing shape variations in biological organisms, allowing for the extraction of morphological features that rely on landmark identification. In avatar creation, matters are more complex. Game designers, when attempting to mimic anticipated ratings, oftentimes cannot rely strictly on landmarks. As we document here, we can relax the rigor of constructed, geometrically identifiable landmarks and replace them with characteristic points identified by artificial neural networks (aNNs), obviating the need for tiresome, meticulous landmark identification by human operators. By integrating these features into the avatar generation process, we can ensure that the resulting avatars exhibit desired characteristics, while still adhering to biological constraints. We relate these characteristic points on female face images whose physical attractiveness has been rated by 50 males and 50 females (on a 7-point scale). We construct Dirichlet distributions of the ratings of each face and use these to investigate whether there are trends or biases in face ratings by male raters versus female raters. We use the coordinates of characteristic points on the faces to identify clusters of attractiveness and relate these clusters to heat maps of the ratings. We explore localized pseudo-symmetries, and how they relate to these ratings. Our proposed system aims to leverage the strengths of transformer architectures and pre-trained neural networks, notably their ability to capture contextual information, to enhance the realism and biological accuracy of the avatars.

Czech name

—

Czech description

—

Type

D - Article in proceedings

CEP classification

—

OECD FORD branch

50101 - Psychology (including human - machine relations)

Project

—

Continuities

N - Vyzkumna aktivita podporovana z neverejnych zdroju

Publication year

2024

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Article name in the collection

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

ISBN

978-3-031-76811-8

ISSN

—

e-ISSN

—

Number of pages

11

Pages from-to

393-403

Publisher name

Springer

Place of publication

Cham

Event location

Washington DC

Event date

Jun 29, 2024

Type of event by nationality

WRD - Celosvětová akce

UT code for WoS article

—