Personality-Aware-Non-verbal-Behavior-Generation

Master’s thesis: a multimodal transformer that generates personality-aware listener avatars (facial expressions, body and hand gestures) in dyadic conversations. Built with PyTorch using VQ-VAE and transformers, trained on UDIVA. Beats a personality-agnostic baseline on FID and P-FID, and 86% of study participants identified the extroverted avatar.

GitHub

Official source

Start locally with the CLI guide