A question about network structure

[JJemaria]

Hi，I found that the network structure in the paper was inconsistent with the code. The 1×1 convolution was used in the feature fusion part, but the code used 3×3 convolution. May I ask which is accurate？

[dys-c]

The 1×1 convolution is used in skip-layer .the code used 3*3 filter in normal process. the two filters is different