Abstract:

This article covers the problem of implementation of trained neural network models in the production environment in order to use them in mobile and web applications. It describes basic frameworks dedicated to the design and training of neural networks, and possible solutions to use during the implementation of a trained model. It also characterizes the training and compilation procedure for a selected convolutional network model trained for the recognition of dog breeds using the Stanford Dogs dataset, and implemented in the production environment with the use of the Flask server and the Docker environment.