Gesture Area Coverage to Assess Gesture Expressiveness and Human-Likeness

Author(s)

Tonoli, Rodolfo Luis ; Paro Costa, Paula Dornhofer ; de Menezes Martins Marques, Leonardo Boulitreau ; Ueda, Lucas Hideki

• Document type: Journal article
• Source: COMPANION PUBLICATION OF THE 26TH INTERNATIONAL CONFERENCE ON MULTIMODAL INTERACTION, ICMI 2024 COMPANION; v. N/A, p. 5-pg., 2024-01-01.

Abstract

This work introduces the analyses of Gesture Area Coverage (GAC) for evaluating the expressiveness of co-speech gestures. GAC explicitly considers the spatial coverage of gestures within motion sequences, an aspect existing metrics neglect. We employ a set of metrics based on GAC to compare motion sequences across two key scenarios. First, to differentiate between distinct gesture styles and a neutral baseline, and second, to assess conditions from the GENEA Challenge 2023, a benchmark for gesture generation systems. In particular, our findings reveal that one of these metrics, the Dice Score, has a stronger correlation with human-likeness ratings compared to the Frechet Gesture Distance. (AU)

• FAPESP's process: 20/09838-0 - BI0S - Brazilian Institute of Data Science
• Grantee: João Marcos Travassos Romano
• Support Opportunities: Research Grants - Research Centers in Engineering Program