Does category imbalance between different categories affect the method?

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )

http://pjreddie.com/darknet/

Other

21.76k stars 7.96k forks source link

Does category imbalance between different categories affect the method? #2506

Open Leontree opened 5 years ago

Leontree commented 5 years ago

Hi, AlexeyAB, About my used dataset, there is a problem of category imbalance between different categories. Does it affect the method? Thanks a looooooooot.

Leontree commented 5 years ago

fig9

AlexeyAB commented 5 years ago

@Leontree Hi,

Usually it isn't a problem since there is used decay: https://github.com/AlexeyAB/darknet/issues/1943#issuecomment-439560675

It reduces the value of frequent features of objects, and gives an advantage to rare features of objects.

But you can try to train with and without focal_loss=1 in each [yolo] layers in cfg-file. And then get the model with the highest mAP.