Brain2Face: A Study on the Relation of EEG Signals and Spatio-temporal Head Avatars

In plain English

The authors ask whether a photoreal 3D head avatar can be animated from brain activity alone, with no camera watching the face. They release a dataset pairing 64-channel EEG recordings with per-frame expression parameters from a personalised 3D Gaussian head avatar, plus a model that maps EEG sequences to those parameters and a benchmark protocol for comparing design choices. This is early-stage work and the usable output is the dataset and protocol rather than a production driving signal, but it is the first serious attempt at camera-free avatar control — which matters for accessibility and for headsets where the face is occluded.