Collection of common code that's shared among different research projects in FAIR computer vision team.
2k
stars
226
forks
source link
Why is the sigmoid function applied to logits before calling F.binary_cross_entropy_with_logits? #126
Closed
kirill-menke closed 1 year ago
In focal_loss.py you can find the following code snipped in
sigmoid_focal_loss
function:https://github.com/facebookresearch/fvcore/blob/fd5043ff8b2e6790f5bd7c9632695c68986cc658/fvcore/nn/focal_loss.py#L35-L36
Why is the sigmoid function applied to the inputs before calculating the loss? Doesn't
F.binary_cross_entropy_with_logits
expect the raw logits?