C++ implementation for facial emotion recognition according to Arriaga's Keras model.
The CNN weights were converted from the pretrained model fer2013_mini_XCEPTION.102-0.66.hdf5
The code was adapted from an example on Caffe repository.
- Ubuntu 16.04
- OpenCV 3.4.2
- Nvidia Caffe
$ ./emotion_recognition.bin --image=path/to/image \
--video=path/to/video \
--net-proto=path/to/prototxt \
--net-weight=path/to/caffeweight \
--net-labels=path/to/labels \
--detector=path/to/haarcascade
Performing emotion recognition using default model
$ ./emotion_recognition.bin --image=image.jpg
Result (with images from FEI face database)