Micro-Expression Recognition with Expression-State Constrained Spatio-Temporal Feature Representations | Synapse