Thanks for your excellent work!
After reading your paper, I benefit a lot. But I have a question about spatial consistency loss.

According to this figure, first calculating the edges in the X, Y , and XY directions, then computing between the pairwise differences of edges in different directions and the zero image. Finally, loss is computed. But the value of zero image is all zero, does it make a contribuation to loss?
And minimizing spatial loss means minimizing the difference between edges in the X and Y, and so on . I think, in different directions, the change of edge information is different, can they be similar?
Looking forward to your reply!