About BLEU scores with default settings
ganymedetitan opened this issue · comments
ganymedetitan commented
Hi! Thank you for opensourcing your qg work.
I tried running with default parameters and because I got errors when running qgevalcap, I evaluated with bleus with coco-caption :
https://github.com/XgDuan/coco-caption/
And got 34.79/19.04/12.29/8.44 for BLEU1~4, which is a bit behind scores reported in ACL paper.
I am not sure did I accidentally do something wrong since I am not familiar with lua codes, or is it just a different implementation/parameter with bleu scores (if there is)?
Sorry for very poor English. :/
Thank you very much!
xinyadu commented
my evaluation scripts are slightly different from the coco-caption scripts,
because they use images as keys, i used the input sentences as keys and
compare the output with *multiple* gold references of the input sentence, i
guess that's why you are getting lower scores.
…On Fri, May 4, 2018 at 3:13 AM, ganymedetitan ***@***.***> wrote:
Hi! Thank you for opensourcing your qg work.
I tried running with default parameters and because I got errors when
running qgevalcap, I evaluated with bleus with coco-caption :
https://github.com/XgDuan/coco-caption/
And got 34.79/19.04/12.29/8.44 for BLEU1~4, which is a bit behind scores
reported in ACL paper.
I am not sure did I accidentally do something wrong since I am not
familiar with lua codes, or is it just a different implementation/parameter
with bleu scores (if there is)?
Sorry for very poor English. :/
Thank you very much!
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#21>, or mute the thread
<https://github.com/notifications/unsubscribe-auth/AJb8YWy9ufesZNkp3brKD914ZqkvvrLpks5tu_-tgaJpZM4TyM7w>
.
ganymedetitan commented
Thank you for quick reply!