# FExGAN-Meta: Facial Expression Generation with Meta Humans

**Authors:** J. Rafid Siddiqui  
**arXiv:** [2203.05975](https://arxiv.org/abs/2203.05975)  
**Date:** 2022-02-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2203.05975v1/

## Key points

- Facial expression recognition has been researched for decades and can now be performed robustly by state-of-the-art methods
- Facial expression generation is a new field that explores the facial expression space, providing a better understanding of underlying semantics
- The resulting system not only provides a labeling system but also tries to map the space and therefore understands the data at a deeper level
- This has many real-world applications, such as video-editing, animation, and game development
- FExGAN-Meta is a proposed method for generating and classifying facial expressions of Meta-Humans that works robustly with images
- Lack of good quality data can hinder deep learning models' performance in recognizing and generating facial expressions
- The FExGAN-Meta model builds on the GAN architecture specifically designed to work with Meta-Humans by mapping their facial expression space
- The proposed method generates facial expressions of Meta-Humans by mapping their facial expression space which allows for exploration into this area providing better understanding into its underlying semantics which can be used in various applications such as video editing, animation or game development
- FExGAN -Meta model shows promising results in generating and classifying images of Meta - Humans' facial expressions which can be used in various applications.

## Summary

Facial expression recognition has been an important subject of research for several decades now and recently it has become a task that can be performed robustly by state-of-the-art methods. However, facial expression generation has emerged as a new field of research that not only deals with the classification problem but also allows exploration of the facial expression space. This is a much more challenging task but provides a better understanding of the underlying semantics of facial expressions. The resulting system not only provides a labeling system but also tries to map the space and therefore understands the data at a deeper level. This has many real-world applications, such as video-editing, animation, and game development. To address this challenge, researchers have proposed a Facial Expression Generation method for Meta-Humans (FExGAN-Meta) that works robustly with images of Meta-Humans. The subtleness of human facial expressions and the large degree of variation in their intensity make it challenging to robustly classify and generate images of facial expressions. Lack of good quality data can hinder the performance of deep learning models. The FExGAN-Meta model was evaluated on a large dataset prepared by collecting images from ten Meta-Humans placed in a studio environment. The results show that FExGAN-Meta robustly generates and classifies simple as well as complex facial expressions exhibited by Meta-Humans. In addition to data, a learning framework requires a robust model. The concept of Generative Adversarial Network (GAN) architecture was proposed to generate images from random Gaussian noise. A classical GAN network consists of two networks: a generator and discriminator network. The generator network takes random samples drawn from Gaussian distribution called latent vectors and reconstructs an image while the discriminant network takes the real image and generated image as input and tries to classify them as real or fake. The FExGAN-Meta model builds on the GAN architecture, but it is specifically designed to work with Meta-Humans. The proposed method generates facial expressions of Meta-Humans by mapping their facial expression space which allows for exploration into this area providing better understanding into its underlying semantics which can be used in various applications such as video editing, animation or game development .The FExGAN -Meta model shows promising results in generating and classifying images of Meta - Humans' facial expressions which can be used in various applications .Overall ,this research contributes to the advancement computer vision research by proposing robust method for generating classifying facial expressions .