search

daijifeng001 / R-FCN

R-FCN: Object Detection via Region-based Fully Convolutional Networks
Other
1.25k stars 543 forks source link

about dilate convolution #28

Closed 1292765944 closed 8 years ago

1292765944 commented 8 years ago

Dear daijifeng: I don't understand why only the res5a_branch2b, res5b_branch2b, res5c_branch2b convolution layer has dilation parameter? what is the point of do that? Thank you!

daijifeng001 commented 8 years ago

Because we removed downsampling between conv4 and conv5, so as to set the output feature stride to 16 instead of 32. Adding dilation can ensure the receptive field of conv5 does not reduce due to the removal.