Facial Expression Recognition in the Wild: Dataset Configurations

Facial Expression Recognition (FER) in the wild has become an increasingly significant and focused area within computer vision, with many studies tackling different aspects to improve its recognition accuracy. This paper utilizes RAF-DB and AffectNet as the two leading datasets in the scene and compares the different experimental dataset configurations to state-of-theart techniques referred to as Amend Representation Module (ARM) and Self-Cure Network (SCN). The paper demonstrates how different dataset configurations should be the main focal point of improving the FER task and how there cannot be significant improvements in the FER task with a lack of a favorable dataset.