An Acceleration Framework for High Resolution Image Synthesis

  • 2019-09-09 03:19:25
  • Jinlin Liu, Yuan Yao, Jianqiang Ren
  • 67

Abstract

Synthesis of high resolution images using Generative Adversarial Networks(GANs) is challenging, which usually requires numbers of high-end graphic cardswith large memory and long time of training. In this paper, we propose atwo-stage framework to accelerate the training process of synthesizing highresolution images. High resolution images are first transformed to small codesvia the trained encoder and decoder networks. The code in latent space is timessmaller than the original high resolution images. Then, we train a codegeneration network to learn the distribution of the latent codes. In this way,the generator only learns to generate small latent codes instead of largeimages. Finally, we decode the generated latent codes to image space via thedecoder networks so as to output the synthesized high resolution images.Experimental results show that the proposed method accelerates the trainingprocess significantly and increases the quality of the generated samples. Theproposed acceleration framework makes it possible to generate high resolutionimages using less training time with limited hardware resource. After using theproposed acceleration method, it takes only 3 days to train a 1024 *1024 imagegenerator on Celeba-HQ dataset using just one NVIDIA P100 graphic card.

 

Quick Read (beta)

loading the full paper ...