Closed xchhuang closed 7 years ago
If you use the probabilistic way of training, you should be using the histogram. Otherwise, you can ignore this example.
Here's the code how we handle such cases when majority mode is used: https://github.com/Microsoft/FERPlus/blob/master/src/ferplus.py#L223-L229
I found that for some images the number of votes on two emotions by the 10 taggers is the same. I wonder how to deal with this problem ?