Ask about the reverse attention module

[tantanhui]

The author writes in the article that A (high-level feature maps) first concatenate B (edge-attention) and then multiply C (expand) in every reverse attention module, but the code shows that A first multiply C and then concatenate B. Which one is the author trying to express?

[DengPingFan]

The author writes in the article that A (high-level feature maps) first concatenate B (edge-attention) and then multiply C (expand) in every reverse attention module, but the code shows that A first multiply C and then concatenate B. Which one is the author trying to express?

Yes, you are right, R = C (f * A, Dow(e_{att})). We will update the formulation accordingly. Thank you very much.

[tantanhui]

There's an error here.
[2020/08/15] Updateing the equation (2) in our manuscript.
R_i = C(f_i, Dow(e_att)) * A_i -> R_i = C(f_i, Dow(e_att)) * A_i;

[DengPingFan]

There's an error here.
[2020/08/15] Updateing the equation (2) in our manuscript.
R_i = C(f_i, Dow(e_att)) * A_i -> R_i = C(f_i, Dow(e_att)) * A_i;

Thanks. It has been fixed.

[tantanhui]

thank you very much!

[1215232494]

Figure 4 also needs to be modified accordingly.