search

chuanqi305 / FocalLoss

Caffe implementation of FAIR paper "Focal Loss for Dense Object Detection" for SSD.
MIT License
171 stars 58 forks source link

About focal loss layer #2

Open zhanglonghao1992 opened 6 years ago

zhanglonghao1992 commented 6 years ago

Hi, I'm wondering have you changed a two-stage net like faster-rcnn or FPN to a one-stage net like the paper said? Do I just need to change the softmax loss layer to your focal loss layer for classification?

chuanqi305 commented 6 years ago

@zhanglonghao1992 Focal loss is not necessary in two-stage detection frameworks. The class imbalance problem only exists in one-stage detection networks such as YOLO and SSD.