Cartoon Face Generation
Personal Project @ CSIE5431 Applied Deep Learning
Personal Project @ CSIE5431
Applied Deep Learning
Generate cartoon face images using Conditional Generative Adversarial Network (Conditional GAN).
Figure 1. Generated cartoon face images.
Proposed by Ian Goodfellow in 2014, Generative Adversarial Networks (GAN) has been applied to various applications, such as image super-resolution, style transfer, image inpainting, etc. In this project, I generated cartoon face images with given face attributes using GAN. For example, given hair color, eye color, face color and glasses, the model will generate a conditional cartoon face. My results are shown above in Figure 1.
Figure 2. The architecture of Auxiliary Classifier GAN (AC-GAN).
I used AC-GAN structure with adversarial loss and auxiliary loss jointly. In addition to the input image X_real and a random vector Z_noise, AC-GAN takes the condition C_class as input of both generator and discriminator and output of discriminator. The architecture of AC-GAN is illustrated above in Figure 2. For the techniques of training GAN, I also did several experiments on model architectures and loss tricks to improve the capabilities of GAN. For more information, please refer to my technical report and my code is also publically available on GitHub.