Focal Loss for Semantic Segmentation

Focal Loss

Focal Loss was proposed by Tsung-Yi Lin et al. in order to improve one-stage object detector. This loss down-weight the loss value of well classified targets. So, this loss allow to detector learn from the hard-example well.

This loss has potential for expansion into other task such as classification or semantic segmentation. In this project, I apply the focal loss to multi-class semantic segmentation.