Closed this issue 4 years ago · 1 comments
Hello @yu4u
May I ask what is the reason of using CrossEntropyLoss instead of MSE/MAE ? From my understanding, age can go beyond 100, and therefore, MSE/MAE loss should be the better criteria for this problem ?
Please refer to https://openaccess.thecvf.com/content_iccv_2015_workshops/w11/papers/Rothe_DEX_Deep_EXpectation_ICCV_2015_paper.pdf