Summary
Focal loss is a type of loss function that downweighs well-predicted labels.
Details
Focal loss () is defined by where is a user-picked modulating factor that decreases the gradient of easy examples as it increases, and is a weight. A value of zero reverts to cross-entropy.
Figures
From https://medium.com/swlh/focal-loss-what-why-and-how-df6735f26616