jshilong / SEPC

Scale-equalizing Pyramid Convolution for object detection(CVPR2020)
Apache License 2.0
330 stars 56 forks source link

How about using Pconv in FPN construction? #8

Closed huangh12 closed 4 years ago

huangh12 commented 4 years ago

Hi~ Thanks for the inpsiring work. From the released code, the Pconv is employed after FPN is contructed, i.e., after p3~p3 are generated. So the Pconv is only used in the retinanet head. I'm curious that how about using pconv during the FPN construction. Have you ever try that?

jshilong commented 4 years ago

Sure, you can try it , we used to remove FPN and only use 4 pconv, I remember just drop <0.5 mAP

huangh12 commented 4 years ago

Thanks for your reply.