search

ruoqianguo / DetNet_pytorch

An implementation of DetNet: A Backbone network for Object Detection.
MIT License
377 stars 117 forks source link

Input Channels for Stage 5 and 6 #18

Open nonobelong opened 6 years ago

nonobelong commented 6 years ago

In the paper it says "our stage 5 and 6 keeps the same input channels as stage 4(256 input channels for bottleneck block)". However in your code the way you create those two stages is take an input of c=1024. I'm actually stuck by this for quite a while. We know that the # of output channels for stage 4 is 1024, then how is it possible that stage 5 takes an input channels of 256?

foralliance commented 5 years ago

@nonobelong use RCNN_toplayer and RCNN_latlayer1