Using a Multi-GPU node to accelerate the training of Pix2Pix neural networks

Abstract Generative adversarial networks are gaining importance in problems such as image conversion, cross-domain translation and fast styling. However, the training of these remains unclear because it often results unexpected behavior caused by non-convergence, model collapse or overly long training, causing task to have be supervised user vary with each dataset. To increase speed Pix2Pix (image-to-image translation) networks, this work incorporates multi-GPU using mixed precision, along optimizations GPU input process. In addition, order make unsupervised terminate when best transformations performed, an early stopping method peak signal noise ratio (PSNR) metric is proposed.