Probability score normalization #2861
Unanswered
johanenforcer
asked this question in
Q&A
Replies: 1 comment 5 replies
-
@johanenforcer yes this is correct. Starting in YOLOv3 the class probabilities are no longer normalized using softmax. This allows for multi-label inference, i.e. one box could have multiple labels like 'person' and 'woman'. |
Beta Was this translation helpful? Give feedback.
5 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
I have noticed that the aggregate of the probability scores retrieved from bounding boxes do not sum up to 1 spefically the x[:, 5:] found in general.py. Can I then use lets say a sigmoid activation function on all of the probability scores for a certain bbox to normalize them for that particular box?
Beta Was this translation helpful? Give feedback.
All reactions