Cannot reproduce the recognition accuracy of certain expressions in the paper

[AMG024]

I used the msceleb weights pre-training model parameters in the paper to train the model’s accuracy on a certain expression that is not as accurate as the recognition accuracy in the confusion matrix in the paper. The parameters used are the default parameters in the paper, so I want to know how In order to reproduce the recognition accuracy of each expression in the confusion matrix of the paper

[nothingeasy]

I meet the same problem . Does anyone reproduce the result in the paper?

[anony-mous-rep]

I also meet the same problem . Is my data set error? I use RAF-DB data set to train, only get the accuracy of 86.44%.

[nothingeasy]

Anyone try this code？ How about the resulut？ Please help~

[yun-kai]

I use RAF-DB data set to train and get the accuracy of 86.25%, and use AffectNet data set to train and get the accuracy of 56.8%.
I don't know why the accuracy of AffectNet is too far from the paper result. Does anyone meet same problem?

[hhwwxx11]

Now the ms-celeb weight pre-training model cannot be downloaded, can you share it?

[lbbbbbbbbbb]

请问你有了么，链接还是失效

[ZechengLi19]

Now the ms-celeb weight pre-training model cannot be downloaded, can you share it?

Same question！

[ZechengLi19]

请问你有了么，链接还是失效

you can check this link https://github.com/zyh-uaiaaaa/relative-uncertainty-learning