Personality-Driven Gaze Animation with Conditional Generative Adversarial Networks
This work addresses the need for more realistic and personality-driven gaze animations in virtual agents, though it is incremental as it builds on existing GAN methods and datasets.
The paper tackles the problem of synthesizing gaze behavior for virtual agents based on personality traits, using a conditional GAN trained on eye-tracking data from 42 participants to generate time series data for gaze targets, blinking, and pupil dimensions, with results applied in a game engine.
We present a generative adversarial learning approach to synthesize gaze behavior of a given personality. We train the model using an existing data set that comprises eye-tracking data and personality traits of 42 participants performing an everyday task. Given the values of Big-Five personality traits (openness, conscientiousness, extroversion, agreeableness, and neuroticism), our model generates time series data consisting of gaze target, blinking times, and pupil dimensions. We use the generated data to synthesize the gaze motion of virtual agents on a game engine.