Why there so many boxes in demo when I trained my model in my dataset after 300 epoches ? anyone can help me ?

VDIGPKU / M2Det

M2Det: A Single-Shot Object Detector based on Multi-Level Feature Pyramid Network

MIT License

1.45k stars 318 forks source link

Open myLeoDK opened 5 years ago

myLeoDK commented 5 years ago

COCO_val2014_000000002154_m2det COCO_val2014_000000005617_m2det COCO_val2014_000000006019_m2det

WIKI2020 commented 5 years ago

这个算法有问题

arc144 commented 4 years ago

Try changing the loss to Focal Loss. I did this and it worked fine training from scratch