In recent years, there has been a surge in the number of deep generative models including Diffusion Models, Generative Adversarial Networks (GANs), and Variational Autoencoders (VAEs). These models are used in various domains such as image, audio, video synthesis, and natural language processing. This paper aims to provide a comprehensive comparison of these models by reviewing their underlying principles, strengths, and weaknesses. GANs generate new data that are similar to a training dataset while VAEs attempt to reconstruct the original high-dimensional input data by mapping this representation back to its original form. Diffusion models gradually add noise to input data to obtain white noise and is not a learnable process. The reverse diffusion process aims to recover the original data and is implemented using a trainable neural network. The paper outlines the key features of different models to guide researchers and practitioners in selecting the most suitable deep generative models for their specific applications.
source update: Comparison of Deep Generative… – Towards AI