Deep Learning for Computer Vision

Many libraries of deep learning algorithms have been written in C++ with a Python interface, but a growing number are also usable with R. In this chapter the authors explore how to program deep neural network models in a few lines of code in order to solve computer vision problems, and they provide some indications on the configurations and parameters allowing their optimization. The two libraries that are now in full expansion are TensorFlow and Torch, or rather PyTorch, which is the Python implementation of Torch and which absorbed the Caffe2 library in March 2018. TensorFlow is currently a reference library in deep learning, known among researchers and developers, for its deep neural network models. The installation of the pure CPU versions of Torch, MXNet, and Keras is immediate, but the installation of their graphics processing units versions requires a prior configuration of the machine, which is not very simple, especially on Windows.