Mastering Computer Vision with TensorFlow 2.x

Neural Network Architecture and Models

The convolutional neural network (CNN) is the most widely used tool in computer vision to classify and detect objects. A CNN maps an input image to an output class or a bounding box by stacking many different layers of linear and nonlinear functions. The linear functions consist of convolution, pooling, fully connected, and softmax layers, whereas the nonlinear layers are the activation functions. A neural network has many different parameters and weight factors that need to be optimized for a given problem set. Stochastic gradient descent and backpropagation are two ways of training the neural network.

In Chapter 4, Deep Learning on Images, you learned some basic coding skills to build and train a ...