You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello, thank you for your job! I have a question need your help.
When |x| >= w, the |x| will subtract the constant C, but I get the constant C is -7.91759 by the formula C = w − w ln(1 + w/epsilon) with w=10 and epsilon = 2. It means that |x| will be larger when |x| >= w. But it not fit the idea of Wing Loss functions, or I am wrong of the constant C when calculating?
Thank you for your help!
The text was updated successfully, but these errors were encountered:
@Dawson-huang Hi, yes, |x| will be larger when |x|>=w due to the shift along the Y-axis with the value of C. But this does not conflict the idea of wing loss. In wing loss, we focus on the gradient of a point, rather than its value. The shift does not change the gradient of a point outside [-w, w]. But the gradient is increased for a point inside [-w, w]. Hope my explanation is helpful for you.
Hello, thank you for your job! I have a question need your help.
When |x| >= w, the |x| will subtract the constant C, but I get the constant C is -7.91759 by the formula C = w − w ln(1 + w/epsilon) with w=10 and epsilon = 2. It means that |x| will be larger when |x| >= w. But it not fit the idea of Wing Loss functions, or I am wrong of the constant C when calculating?
Thank you for your help!
The text was updated successfully, but these errors were encountered: