A littel difference between the figure in the paper and the code.

[GuideWsp]

the upper branch has no convolution in the paper, while the code do.

[njulj]

This convolution has little compact to the final performance, so you can safely remove it. While calculating the parameters of our model, we use this convolution.

[GuideWsp]

Thanks for you reply, i do think it has little influence to the final performance.

[shanwua]

When will you open source your code？

[star-lunar]

Hello, thank you for taking time to read it.

May I ask how the graph Figure8 Average feature maps of residual blocks used in the paper was drawn?

I would be very appreciate that if you could answer the question and give an early reply