A systematic literature review of generative adversarial networks (GANs) in 3D avatar reconstruction from 2D images

The rapid advancement of machine learning and computer vision has paved the way for significant processes in 3D avatar reconstruction from 2D images. Recently, generative adversarial networks (GANs) have emerged as a promising approach for generating realistic and detailed 3D avatar from a 2D images. This systematic literature review provides a comprehensive overview of research on 3D avatar reconstruction from a 2D image using GANs. 35 relevant studies from 2014 to 2022 using the databases ACM, IEEE Xplore, ScienceDirect and Web of Science were identified and analyzed. The review covers a wide range of topics, including network architectures, training methodologies, performance and evaluation metrics employed in the context of 3D avatar reconstruction using GANs. The following research questions are addressed: RQ1) What types of GAN models are employed in reconstructing 3D avatars from 2D images? RQ2) How is their performance in 3D avatar reconstruction? RQ3) What are the limitations and future directions of GAN models in 3D avatar reconstruction from 2D images? The findings indicate significant progress in generating high-fidelity 3D avatars using GAN-based approaches. Various GAN architectures are employed, each offering unique advantages and limitations. The identified challenges include the accuracy of fine-grained details, robustness to lighting and limited diversity of the training dataset. Future research directions involve improving accuracy and realism via enhanced details. In conclusion, this systematic literature review provides a comprehensive understanding of the state-of-the-art algorithms and methodologies in 3D avatar reconstruction using GANs. The review identifies research gaps and suggests potential directions for future investigation.