2017/04/07 by Dieu Linh Tran, Tran, Dieu Linh, Robert Walecki +9 · 2 citations
Computer Science · Psychology · #Computer Vision and Pattern Recognition (cs.CV) #Emotion and Mood Recognition #FOS: Computer and information sciences #Face recognition and analysis #Generative Adversarial Networks and Image Synthesis
paper · pdf · doi:10.48550/arxiv.1704.02206
openalex publication_date 2017/04/07 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Human face exhibits an inherent hierarchy in its representations (i.e.,\nholistic facial expressions can be encoded via a set of facial action units\n(AUs) and their intensity). Variational (deep) auto-encoders (VAE) have shown\ngreat results in unsupervised extraction of hierarchical latent representations\nfrom large amounts of image data, while being robust to noise and other\nundesired artifacts. Potentially, this makes VAEs a suitable approach for\nlearning facial features for AU intensity estimation. Yet, most existing\nVAE-based methods apply classifiers learned separately from the encoded\nfeatures. By contrast, the non-parametric (probabilistic) approaches, such as\nGaussian Processes (GPs), typically outperform their parametric counterparts,\nbut cannot deal easily with large amounts of data. To this end, we propose a\nnovel VAE semi-parametric modeling framework, named DeepCoder, which combines\nthe modeling power of parametric (convolutional) and nonparametric (ordinal\nGPs) VAEs, for joint learning of (1) latent representations at multiple levels\nin a task hierarchy1, and (2) classification of multiple ordinal outputs. We\nshow on benchmark datasets for AU intensity estimation that the proposed\nDeepCoder outperforms the state-of-the-art approaches, and related VAEs and\ndeep learning models.\n