Projects Last Projects

The main goal is to generate synthetic data for projects, in the machine learning field, that deals with face emotions classification.
To classify images with multiple class labels using only a small number of labeled examples is a difficult task. Especially when the label (class) distribution is imbalanced. In face emotion classification we have imbalanced label distribution because some classes of emotions are relatively rare comparing to others. For example, disgust emotion is more rare than happy or sad.
In this work, we propose a data augmentation method using generative adversarial networks (GAN). It can complement and complete the data manifold and find better margins between neighboring classes. Specifically, for this task, we are using classifiers based on Convolutional Neural Networks (CNN) and a variation of cycle-consistent adversarial networks such as CycleGAN, Improved CycleGAN, and The Wasserstein CycleGAN. The CycleGAN is a direct implementation of Emotion Classification with Data Augmentation Using Generative Adversarial Networks paper.
In order to improve the results and avoid problems that we faced we employ different variations of CycleGANs. We show, that our empirical results can obtain a ~5% increase in the classification accuracy, after employing the GAN-based data augmentation techniques.

Please, see project papers.

Please, see project report.

Please, see final presentation.