Generating Images Part by Part with Composite Generative Adversarial Networks

TL;DR

This paper introduces composite GANs that generate images part by part using multiple generators, combining their outputs through alpha blending, enabling unsupervised learning of complex image structures.

Contribution

The paper presents a novel composite GAN architecture with multiple generators that learn to generate different image parts without supervision.

Findings

Successfully generated images with distinct parts like background and face

Demonstrated the model's ability to learn image structure empirically

Unsupervised training without labeled data

Abstract

Image generation remains a fundamental problem in artificial intelligence in general and deep learning in specific. The generative adversarial network (GAN) was successful in generating high quality samples of natural images. We propose a model called composite generative adversarial network, that reveals the complex structure of images with multiple generators in which each generator generates some part of the image. Those parts are combined by alpha blending process to create a new single image. It can generate, for example, background and face sequentially with two generators, after training on face dataset. Training was done in an unsupervised way without any labels about what each generator should generate. We found possibilities of learning the structure by using this generative model empirically.