Loss Functions for Object Detection
Paper:
Code:
Paper:
Code:
Binary Cross-Entropy
Weighted Binary Cross-Entropy
β is utilized to control amount of false negatives and false positives.
(β > 1) for reducing the false negative.
(β < 1) for reducing false negative.
Balanced Cross-Entropy
β = 1 - y/(H*W) with H*W = total pixels.
Focal Loss
Limit the effect of easy samples on loss functiunStrongly punish
https://github.com/implus/GFocal