Questions about part of the network input ΔF_s

[DaddyJin]

Thank you for your great work! But I am confused about the input ΔF_s.
In the paper, each f_i(WH2) in ΔF_s is a spatial field map indicating the landmark difference.
Does it mean that each point in f_i is a coordinate, in the position of s_i is the value of Δs_i and other position is (0, 0)?
And how ΔF_s(WH2) can be shown as a facial landmark image in Figure 1?
Thank you for your reply and looking forward to the code!

[kgu3]

Hi Jin, For your 1st question: you are right. For each image in the input image bank, f_i is a WxHx2 tensor, where WH are image size, and 2 because there are x and y coordinates. Each location in f_i (for example f_i(x,y,0)) is either 0 (when (x,y) is not one of the 68 landmarks), or the difference between target and source landmark (in the case of f_i(x,y,0), it is the value of the difference between x coordinates). In other words, we compute the 68 landmark difference, and fill-in the values into the corresponding locations in f_i. For 2nd question: The image of landmark input in Figure 1 is just a demonstration. Strictly speaking, we should show a arrow field where only 68 arrows will have non-zero length on a W*H image. We thought that the arrow map would be hard to see, so just decided to use the current version. Sorry about the confusion. For the code, Really apologize for the prolonged delay. For one thing we are trying to apply for patenting, which might delay the code release. In addition, I'm quite entangled by my TA and graduation recently. Regards, Kevin DaddyJin <notifications@github.com> 于2020年3月28日周六 下午9:59写道：

…

Thank you for your great work! But I am confused about the input ΔF_s. In the paper, each f_i(W*H*2) in ΔF_s is a spatial field map indicating the landmark difference. Does it mean that each point in f_i is a coordinate, in the position of s_i is the value of Δs_i and other position is (0, 0)? And how ΔF_s(W*H*2) can be shown as a facial landmark image in Figure 1? Thank you for your reply and looking forward to the code! — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEE42POZVECI5B5JT6SQ6M3RJ22PLANCNFSM4LVZYSIA> .