Challenge report: Compound Emotion challenge

This paper is a brief introduction to our submission to the Joint Challenge on Compound Emotion Recognition and Multhnodal (Audio, Facial, and Gesture) based Emotion Recognition (CER&MMER) held in conjunction with the IEEE International Conference on Automatic Face and Gesture Recognition (FG) 2020. Our method combine texture information and landmark, which is the geometric representation of emotions as the input of the model. Moreover, we proposed a neHvork based on the Alexnet network. To better deal with the recognition of micro emotions, we combine cross-entropy loss and triplet loss as our loss function. Finally, we get better performance than the baseline on the test set.