Course IV (convolutional neural Networks), fourth week (special Applications:face recognition & Neural style transfer)--1.practice Quentions

Explain

This allows us to learn to predict a person ' s identity using a Softmax output unit, where the number of classes equals the Number of persons in the database plus 1 (for the final "not in Database" Class).

Reasons for the above options error:

1, plus 1 explanation error:

Put someone's photo into the convolutional neural network, use the Softmax unit to output the kind, or label, to correspond to these different people, or not any of them, so in Softmax we have more than one kind of output. But this is actually not good, because such a small training set is not enough to train a robust neural network.

2. For the final ' not in database ' class statement error

In a learning problem, only one sample can be learned to be able to recognize the same person. Most face recognition systems need to solve this problem because each employee or team member in your database may have only one photo, not a photo of the person.

Explain

Given a content image C, given a style picture of s, and the goal is to generate a new picture of G. To achieve a neural-style migration, it is necessary to define a cost function J for G to evaluate the quality of a generated image, and we will use the gradient descent method to minimize J (g) in order to generate the image. Instead of simply entering two pictures, train the network to create a new composite image.

Explain

Output of the volume f= (32+2*0-3)/1 + 1 = 30, 32 filters, so the output volume dimension is 30*30*30*32

Course IV (convolutional neural Networks), fourth week (special Applications:face recognition & Neural style transfer)--1.practice Quentions