Original subtitles

1

Hello and welcome to this Python tutorial we have just defined in the previous Statoil the architecture

2

of the neural network of the generator.

3

And now we need to add a tool to our class G which will be the forward function that will forward propagate

4

the signal inside the whole neural network.

5

The good news is that after the storm we had in the British Statoil.

6

Now things are going to get really easy because we will simply use the main object which remains an

7

object of the sequential class that contains all the different modules to forward propagate the signal

8

in just one line of code.

9

Let's do it.

10

We are going to define a new function.

11

There are forms starting with death and then we need to specify the name of this function which will

12

be forward for the forward propagation and this function is going to take two arguments.

13

Self our object which represents nothing else than the neural network itself.

14

The neural network of the generator.

15

And also because as I've just said we're going to use the main object to propagate the signal but since

16

this main object is a variable attached to our object.

17

Well we need to input ourselves object here to be able to use self-determine.

18

All right.

19

And then the second argument you can see the year is actually going to be the input that is nothing

20

else than the input of the neural network of the generator.

21

I remind this input is going to be some random vector of size 100.

22

That's why we specified 100 here for the size of the input of our first inverse convolution.

23

This is going to be a random vector input of the generator which will just represent some noise to generate

24

a fake image and that fake image will be the output of the generator.

25

It will be the fake image generated by the generator.

26

But to get this fake image we need to create this random vector noise that will be exactly this input.

27

The input of the generator and this input will have a size of 100.

28

It will become much more clear when we start the training of our two brains.

29

The brain of the generator and the brain of the discriminator because right now you only see some names

30

for the variable such as inputs but you'll see that we will create those actual inputs later on in the

31

training phase.

32

So no worries if it's not 100 percent crystal clear now it will become much clearer when we start the

33

training face for now.

34

Just make sure to understand that we're creating the generator and the discriminator.

35

And right now we're making the forward function that will forward propagate the signal inside the neural

36

network at the generator.

37

All right so now things are going to get super easy.

38

We are going to take our self duct main memory module which is the metal module of our neural network

39

object represented by self.

40

We're going to feed this neural network with the inputs which so far is just an argument but then will

41

become noise random vector we are fitting the new one that work with the input and this will return

42

the output because the signal is propagated through all the different modules layers of this neural

43

network and then in the end we'll get the output which of the three channels of the fake generated images.

44

And therefore since we get now here I am introducing a new variable that I'm calling output and that

45

is equal to exactly what is returned by the main metor module.

46

All right so now that we get the output which just need to return it and therefore I'm just entering

47

the fourth function with return output.

48

And there we go we have our output of the generator.

49

Perfect.

50

So double congratulations because you made the forward function and because you're done with the whole

51

class defining the generator but be careful we just defined the class.

52

Nothing is created yet.

53

And so in the next to toile we'll have to create an instance that is not checked of the class of the

54

generator class.

55

And this object will be nothing else than the neural network of the generator defined by the following

56

properties and containing the tool or the Ford function to propagate the signal inside the neural network.

57

So let's create this object in the next tutorial.

58

It will be quick and easy will just create the object and we'll just apply the weights in it function

59

to initialize the weights of the neural network as it should be.

60

That is according to the conventions of the adversarial networks.

61

So let's do that in the next tutorial.

62

And until then enjoy computer vision.