why i train own dataset ，AP is lower than 0.1?

tianzhi0549 / FCOS

FCOS: Fully Convolutional One-Stage Object Detection (ICCV'19)

https://arxiv.org/abs/1904.01355

Other

3.27k stars 630 forks source link

why i train own dataset ，AP is lower than 0.1? #278

Open 15241192958 opened 4 years ago

15241192958 commented 4 years ago

i train the model with my own dataset , just as coco type with the fcos_R_50_FPN_1x.yaml ,the loss is 1.38 ,after 4hours,it stop ,when i evaluate the output model ,the AP is just 0.09.please tell me how to improve? thank u

cunjif commented 4 years ago

如果修改了网络层,检测网络层的名字是否带有module, 如果有,就改掉,再试试

cunjif commented 4 years ago

@15241192958