This article discusses how to speed up GAN training by using TPUs in Kaggle or Google Colab. The author provides a step-by-step guide and suggests using a dataset with 200,000 full-color images of famous people’s faces for the purposes of the article. The article includes code and functions for loading and preparing the dataset, connecting to TPUs, and normalizing the pixel values of the images. The author also explores the use of the MTCNN face detection library to select the part of the image that contains the face for better GAN training.

source update: How To Use TPUs in Kaggle / Google Colab To Train a GAN in the… – Towards AI