This is my personal exercises that use fer2013 datasets with Tensorflow to classify face emotion. I used vgg and resnet respectively,but all perform poor.In the test datasets,two model achieve accuracy of 54% and 48%.
Except the reasons I do incorrectly, there are several problems with this datasets:
- Pixel is too low
- Have a front face, side face, cartoon face and so on. Some kinds of faces that I can't tell what.
- Data is not a face.
The pre-trained model be put into model/ folder.
Vgg:
Loss
Accuracy
Resnet:
Loss
Accuracy
