Image Recognition
The authors of this review paper [1] pay attention to the third topic, ID-invariant FER. Then, the (facial) expression tendency between IDs is quantified by calculating the difference between these ID sets packed by ID. At this time, an optimal transport (OT) problem that maps elements of two (discrete) sets to minimize a given cost function is utilized. The $\mu$ value obtained in this way shows how different the emotional representation is between ID sets. We expect to be able to learn the ID-invariant representation by performing supervision learning with GT through this normalized feature.