Please use this identifier to cite or link to this item:
|Title:||Generative neural network for emotion recognition||Authors:||Chen, Hailin||Keywords:||DRNTU::Engineering::Computer science and engineering||Issue Date:||2019||Abstract:||Recognising affect from visual data has long been a research interest. However, annota- tions of affect in images/videos are expensive to acquire and current datasets all have limitations of either being too small or containing imbalanced affect classes. . In other at- tempts of data augmentation for emotion recognition, generation is all modelled as image translation task. In this paper, we first analyse generative models and multiple relevant GAN variants. We then propose to boost performance of emotion recognition model by investigating two generative models with one being image translation model using GANs and the other model to generate target data distribution with latent noise as input. In this way, we can achieve richer and more flexible data augmentation. Experiments on fer2013 dataset showed effectiveness of our methods.||URI:||http://hdl.handle.net/10356/77200||Rights:||Nanyang Technological University||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SCSE Student Reports (FYP/IA/PA/PI)|
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.