search

EricGuo5513 / action-to-motion

Official implementations for "Action2Motion: Conditioned Generation of 3D Human Motions (ACM MultiMedia 2020)"
MIT License
159 stars 21 forks source link

Wrong FID calclulation #7

Closed atnikos closed 2 years ago

atnikos commented 2 years ago

Hey, thanks for the nice work! One thing I have observed is that the FID calculation is wrong: here you are passing to the function (gt, generated) features for FID calculation. However the frechet distance function accepts the arguments in the reversed order which if you do results in different number.

EricGuo5513 commented 2 years ago

Hi, sorry for the late reply. Thanks for pointing it out. That does make some differences.

atnikos commented 2 years ago

Hey, I don't think it is a problem in your case this problem occurs if you normalize the features before you apply frechet distance, which I don't think you do. Otherwise, Frechet Distance is commutative and does not cause any difference in the results.