Zheng222 / DMFN

Image Fine-grained Inpainting (Winner Award of ECCVW AIM 2020 Extreme Inpainting Track1&Track2)
248 stars 29 forks source link

A question about align loss #4

Closed HannH closed 4 years ago

HannH commented 4 years ago

Hello, I have a question about Equation 6,7 in your paper. In Equation 6, we get c_u_k by cumulating R along axis u,v , So why can we cumuylate along axis u,v again in Equation 7? In my opinion, the shape of c_u_k is (B,k), right?

Zheng222 commented 4 years ago

@HannH Hello, in Eq. 6, the c_u_k is coordinate expectation along axis u. The definition of expectation is the sum(variable * probability distribution function). The shape of c_u_k is (B, 1).

HannH commented 4 years ago

@HannH Hello, in Eq. 6, the c_u_k is coordinate expectation along axis u. The definition of expectation is the sum(variable * probability distribution function). The shape of c_u_k is (B, 1).

Thanks, but I still can not understand the symbol sigma(u,v) in Equation 7 since you said the shape of c_u_k is (B,1), I mean there is even no axis u or axis v in c_u_k to accumulate, right?

Zheng222 commented 4 years ago

@HannH The updated equation 7 is shown as follows image

HannH commented 4 years ago

Thanks for your reply