Synthetic generation of photorealistic 3D characters for face-based episodic memory measures
Year: 2025
Type: preprint
Abstract: In Episodic Memory (EM) research, face identification and association tasks have emerged as sensitive indicators of memory decline. While 2D images have been widely used in face-based tasks, 3D representations offer unique advantages that can enhance ecological validity of stimuli, by incorporating dynamic behaviours that more closely resemble real-world conditions, such as variations in pose, facial expressions, and movements. Generative Adversarial Networks (GAN) can be a powerful tool for creating photorealistic facial stimuli for these experiments. Unlike traditional methods, GAN can allow researchers to systematically control a variety of facial features, while preserving each model's identity. This article introduces a novel approach to creating stimuli for EM and other face-based cognitive research, combining GAN-based photorealistic face generation with seamless 3D full-body character integration. This method provides parametric control over head pose and facial gesture, along with age, sex, racial attributes. Crucially, this approach introduces a systematic quantification and parametric control over the perceptual differences between facial identities, that bridges a critical methodological gap in face-based cognitive research. Results from a validation test with 313 volunteers provides compelling evidence for the efficacy of this method and the validity of the generated stimuli.
Authors: Alvaro Pastor, Pierre Bourdin Kreitz
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: gold
OpenAlex ID: W4413060051