cls_normalizer and iou_normalizer

Zzh-tju / DIoU-darknet

Distance-IoU Loss into YOLO v3

GNU General Public License v3.0

313 stars 85 forks source link

Open chenyang9799 opened 4 years ago

chenyang9799 commented 4 years ago

cls_normalizer and iou_normalizer对模型有什么影响？为什么要设置为1和0.5？论文里面没有提及啊。谢谢

Zzh-tju commented 4 years ago

readme里有介绍，这是损失的权重系数。前者是分类损失，后者是bbox regression损失。顺便一提，这个系数对不同的数据集也会不同