I'm trying to reimplement your network but having problem on the DSCFA module. In the paper you wrote "Our DSCFA block learns dilated spatial features to extract and aggregate dilated contextual features from four directions" and the figure showed in Figure 7.
I'm wondering:
1) how to take the convolution from different directions? Or they are just the output for the normal convolution with different given meanings?
2) what are the shapes of the kernels you use to get the DSCNN_U/D/R/L?
Hello,
I'm trying to reimplement your network but having problem on the DSCFA module. In the paper you wrote "Our DSCFA block learns dilated spatial features to extract and aggregate dilated contextual features from four directions" and the figure showed in Figure 7. I'm wondering: 1) how to take the convolution from different directions? Or they are just the output for the normal convolution with different given meanings? 2) what are the shapes of the kernels you use to get the DSCNN_U/D/R/L?
Thanks!