REDES NEURAIS CONVOLUCIONAIS PARA RECONHECIMENTO FACIAL EM PHP

This paper goal is to compare an implementation of convolutional neural networks in both PHP and Python, expliciting their respectives performances in facial recognition. For PHP, the library Rindow Neural Networks was chosen for the project, meanwhile in Python, Keras using Tensorflow was the choice. For the experiments, both languages follow the same structure for the neural network, using the same dataset, compromised of 16 classes. Both codes were inserted in a web frameworks, Slim for PHP and Flask for Python. The results show that the neural network built in PHP has a higher average accuracy (0,954) comparedd to the Python network (0,860), despite having a higher average training time. In the end, PHP proved to be a viable alternative in specific situations, namely when dealing with legacy projects and codebases. Despite that, Python with its bigger versatility and community support, is still the recommended option for machine learning projects and for learning enviroments, thanks to its ease of use and good documentation. Despite the viability of PHP for building neural networks, Python is still the most practical and efficient choice for most scenarios. For future works an study of the Rubix library, another machine learning tool for PHP, could fix some of Rindow shortcomings.