3D Human Body Models: Parametric and Generative Methods Review

This paper provides an overview of the current state-of-the-art in the field of 3D human body model estimation, reconstruction, and generation in computer vision. The paper focuses on the most widely used parametric and generative methods and their applications. The paper highlights the use of different input data formats, including 2D images, videos, and 3D scans of the human body in various fields such as medicine, film industry, video game industry, extended reality, and clothing. The field of 3D human body recovery has seen a significant advancement with the development of parametric models. These approaches use a set of parameters to represent body shape and pose and are widely used for reconstructing 3D human body. Some approaches emphasize body deformations, while others concentrate on shape and pose optimization or the separation of body shape into identity-specific and pose-dependent components, among other aspects. The advancements in the field have led to improved accuracy and stability in representing human body shapes and poses. On the other hand, in recent years various generative methods have been developed to generate 3D models of the human body. Variational Autoencoder (VAEs) and Generative Adversarial Networks (GANs) are two commonly used types of neural networks for this purpose. These methods can generate 3D human body models by learning the distribution of the data.