Face Generation using DCGAN for Low Computing Resources

Image generation is the task of generating brand new images from existing datasets. With the increasing development of the Generative Adversarial Network (GAN), it is now widely used in the field of image generation. Research based on its time cost and the convergence of its training process has been widely concerned. However, existing techniques for network convergence during the training process usually require high computing resources. This paper has used Deep Convolutional Generative Adversarial Network (DCGAN) on the CelebA dataset for experimental evaluation: The number of training epochs and the algorithm’s parameters is adjusted to achieve a balance between the quality of generation results and computing resources used. This is assumed to be convenient for future studies under the condition of limited computing resources. Besides, by adjusting the parameters of the optimization algorithm, the convergence of GAN under the conditions of a specific optimization algorithm is studied.