Deep learning-based evaluation system for assessing actor performance using facial images

In cinema, audiences are often mesmerized by the performance of actors/actresses due to their seamless acting skills. Despite enacting every expression aptly, sometimes these performers have to face bias in the industry, be it nepotism, unfair opportunities or any other factors. In this research article, we are proposing a model built on the foundations of Convolutional Neural Network (CNN) and Vision Transformer (ViT) for analyzing the performance of actors based on their facial expressions for a particular scenario in an unbiased manner. The architecture of the proposed model includes a convolution layer, max pooling, GeLU activation function, augmentation, batch normalization and an intermediate Vision Transformer. Two open-access datasets CK + and KDEF have been employed for training and testing of the proposed model CNN_ViT, to achieve an accuracy of 97.99% and 96.59% on the aforementioned datasets. The results indicate that the proposed model, CNN_ViT, outperformed other machine learning models in terms of accuracy while evaluating the performance.