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

Kód výsledku v 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>

Výsledek na webu

<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>

Jazyk výsledku

angličtina

Název v původním jazyce

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

Popis výsledku v původním jazyce

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.

Název v anglickém jazyce

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

Popis výsledku anglicky

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.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

50101 - Psychology (including human - machine relations)

Projekt

—

Návaznosti

N - Vyzkumna aktivita podporovana z neverejnych zdroju

Rok uplatnění

2024

Kód důvěrnosti údajů

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

Název statě ve sborníku

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

—

Počet stran výsledku

11

Strana od-do

393-403

Název nakladatele

Springer

Místo vydání

Cham

Místo konání akce

Washington DC

Datum konání akce

29. 6. 2024

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—