Spatio-temporal convolutional emotional attention network for spotting macro- and micro-expression intervals in long video sequences

•The STCEAN model considers the changes of spatial features in the temporal dimension.•The MAS of two heads is focus on different emotional dimensions attention weight.•The STCEAN model uses focal loss function to reduce sample imbalance.•Leave-Half-Subject-Out (LHSO) cross-validation method to reduce trained time.