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
When testing CIDEr with one reference sentence for each candidate sentence, I get a CIDEr score of 7.24. Since CIDEr computes the average cosine similarity between a candidate and a reference, is it normal for CIDEr values to be above 1? Thank you!
The text was updated successfully, but these errors were encountered:
Yes, it is normal for CIDEr-D values to be greater than 1, infact the maximum value is 10, because the cosine similarity is multiplied by 10 (See Eqn. 4 in https://arxiv.org/pdf/1411.5726.pdf).
Why are both the CIDEr score and the CIDEr-D score implementations multiplied by 10 in the end? I can get both metrics above one. Is there any implementation error for the original CIDEr score?
When testing CIDEr with one reference sentence for each candidate sentence, I get a CIDEr score of 7.24. Since CIDEr computes the average cosine similarity between a candidate and a reference, is it normal for CIDEr values to be above 1? Thank you!
The text was updated successfully, but these errors were encountered: