# EmotioNet Challenge: Recognition of facial expressions of emotion in the wild

**Authors:** C. Fabian Benitez-Quiroz, Ramprakash Srinivasan, Qianli Feng, Yan Wang, Aleix M. Martinez  
**arXiv:** [1703.01210](https://arxiv.org/abs/1703.01210)  
**Date:** 2017-03-03  
**Summary page:** https://www.summarizepaper.com/en/paper/1703.01210v1/

## Key points

- EmotioNet challenge evaluated computer vision algorithms in analyzing facial expressions of emotion in real-world images
- Two tracks: detecting 11 action units (AUs) and recognizing 16 basic and compound emotion categories
- Current algorithms struggle to reliably solve these tasks, with greater limitations in recognizing emotions
- Algorithms unaffected by mild resolution changes, small occluders, gender or age
- 3D pose identified as a significant limiting factor for performance
- Areas that require special attention for future advancements discussed
- Challenges faced by existing algorithms in accurately analyzing facial expressions of emotion highlighted
- Potential avenues for improvement suggested

## Summary

This paper presents the methodology and findings of the EmotioNet challenge, which aimed to evaluate the performance of computer vision algorithms in analyzing facial expressions of emotion in real-world images. The challenge consisted of two tracks: one focused on detecting action units (AUs), specifically 11 AUs, while the other track aimed to recognize 16 basic and compound emotion categories. The results indicate that current computer vision and machine learning algorithms struggle to reliably solve these tasks, with greater limitations observed in recognizing emotions. Interestingly, the algorithms were found to be unaffected by mild resolution changes, small occluders, gender or age. However, 3D pose was identified as a significant limiting factor for performance. The paper concludes with a comprehensive discussion on areas that require special attention for future advancements in this field. It also sheds light on the challenges faced by existing algorithms in accurately analyzing facial expressions of emotion and highlights potential avenues for improvement.