Visualizing the Invisible: Generative Visual Grounding Empowers Universal EEG Understanding in MLLMs

In three linesGVG (Generative Visual Grounding) uses an EEG-to-image generative model to translate brain activity into visual images, bypassing text-only alignment. Tested on GVG-X-Omni (170M tuned params) and GVG-Janus (trimodal), the framework improves EEG understanding and visual generation by leveraging MLLMs' visual priors.

Summary generated by Claude — human-verified