Buggy implementation of smooth L1 loss
Closed this issue · 1 comments
flovst commented
The implementation in pytorch/model/detector/regression_loss (line 144)
seems to apply x (instead of x_abs) in the case selection.
Simply changing is_negative_branch = (x < (1.0 / sigma_squared))
to is_negative_branch = (x_abs ...).
shall fix it.
Thank you for this project!
trzy commented
Good eye, thank you! Fixed.