STAG Network for Spatio-temporal Micro-expression Recognition

STAG uses a dynamic ROI-AU-coupled spatial-temporal network to recognize micro-expressions by jointly modeling motion flow and facial connectivity. The architecture integrates a graph attention network for spatial reasoning with a transformer encoder for temporal modeling to capture subtle facial muscle activations.