An Ensemble of VGG Networks for Video-Based Facial Expression Recognition

This paper presents a fusion-based ensemble of VGG networks for the Multimodal Emotion Recognition Challenge 2017. Image fusion is used to aggregate consecutive frames from video sequences for the representation of temporal information. Then, an ensemble of four VGGFace models which have been fine-tuned on the MEC dataset, is utilized to extract facial expression features from the fused images. The VGGFace-BiLSTM and VGGFace-Bi-GRU arealso implemented for comparison. The accuracies of the fine-tuned VGGFace-ensemble, VGGFace-Bi-LSTM and VGGFace-Bi-GRU on validation data are 51.06%, 43.95% and 44.92% respectively, indicating the effectiveness of our method.