Thesis Project: A multimodal transformer-based generative model that creates listener avatars conditioned on personality traits to produce realistic non-verbal responses (facial expressions, body and hand gestures), during dyadic conversations. Built with PyTorch, and trained on UDIVA dataset, achieving state-of-the-art FID/P-FID performance.
computer-vision generative-model avatar-generator human-computer-interaction cognitive-robotics personality-traits ai-agents multimodal-deep-learning transformer-encoder gesture-generation pose-detection vision-transformer large-vision-language-models vector-quantized-variational-autoencoder personality-aware-models
-
Updated
Jan 5, 2026 - Python