Recent advances on loss functions in deep learning for computer vision

Yingjie Tian, Duo Su, Stanislao Lauria, Xiaohui Liu

Abstract: Highlights•The properties and comparisons about each loss function are presented.•Two guidelines on designing or selecting the loss functions are proposed.•The application scenarios about each loss function are summarized.•The SOTA results of each task and the losses used in SOTA are summarized.•Some frontier challenges are also discussed in our review.