E-CNN ground truth

[yutseng318]

Hi,

As you have mentioned in your paper 4-1, the sub-networks are trained separately. I just wonder about how you obtain the label of your E-CNN, which should be an edge map(and denoted E^t* in your paper).

[fqnchina]

The label of E-CNN is the edge map of ground truth target image. Note the edge map is not the one generated by traditional edge detector, it is computed by our edge computation layer, which is basically like the gradients.

[yutseng318]

Ok, tkx for the reply!