Multi-class Generative Adversarial Nets for Semi-supervised Image Classification
This addresses a specific bottleneck in semi-supervised image classification for computer vision applications, but it is incremental as it modifies existing GAN training.
The paper tackled the problem of GANs generalizing too much when classifying similar image classes in semi-supervised learning, leading to poor performance, and proposed a modified training method that improved multi-class classification on MNIST and Fashion-MNIST datasets.
From generating never-before-seen images to domain adaptation, applications of Generative Adversarial Networks (GANs) spread wide in the domain of vision and graphics problems. With the remarkable ability of GANs in learning the distribution and generating images of a particular class, they can be used for semi-supervised classification tasks. However, the problem is that if two classes of images share similar characteristics, the GAN might learn to generalize and hinder the classification of the two classes. In this paper, we use various images from MNIST and Fashion-MNIST datasets to illustrate how similar images cause the GAN to generalize, leading to the poor classification of images. We propose a modification to the traditional training of GANs that allows for improved multi-class classification in similar classes of images in a semi-supervised learning framework.