A pre-trained model vs dedicated convolution neural networks for emotion recognition

Facial expression recognition (FER) is one of the most important methods influencing human-machine interaction (HMI). In this paper, a comparison was made between two models, model that built from scratch and trained on FER dataset only, previously data set containing various images, which VGG16 model, then reset using dataset. The FER+ augmented to be used in training phases proposed models. models will evaluated (extra validation) by images internet order find best for identifying human emotions, where Dlib detector OpenCV libraries are face detection. results showed emotion convolutional neural networks (ERCNN) dedicated emotions significantly outperformed pre-trained terms accuracy, speed, performance, 87.133% public test 82.648% private test. While it 71.685% 67.338% model.