search

microsoft / FERPlus

This is the FER+ new label annotations for the Emotion FER dataset.
https://arxiv.org/abs/1608.01041
Other
601 stars 169 forks source link

the loss function for multi label learning #8

Open amirhfarzaneh opened 6 years ago

amirhfarzaneh commented 6 years ago

In your paper you have defined a particular loss function in equation (4). I wanted to know where this loss function is derived from? Is there a source for it or is it your contribution?

Also when you're implementing this in your code: https://github.com/Microsoft/FERPlus/blob/master/src/train.py#L44

you're calculating the logarithm after finding the maximum not on the predictions which is shown as log(q) in equation(4). Is there a particular reason for this? I would appreciate more elaboration