arxivst stuff from arxiv that you should probably bookmark

Spatiotemporal Networks for Video Emotion Recognition

Abstract · Apr 3, 2017 13:21 ·


Arxiv Abstract

  • Lijie Fan
  • Yunjie Ke

Our article presents an audio-visual based multi-modal emotion classification system. Considering the fact of deep learning approaches to facial analysis have recently demonstrated high performance, in our work, we use convolutional neural networks (CNNs) for emotion recognition in video, relying on temporal averaging and pooling operations reminiscent of widely used approaches for the spatial aggregation of information. In respect of time sequence, we extract the feature from audio clips in the video and use RNN to propagate information. In this work, we focus our presentation and experimental analysis on a fusion CNN-RNN architecture for facial expression analysis.

Read the paper (pdf) »