Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated: 1 00:00:00,210 --> 00:00:02,760 Hello and welcome to this new tutorial. 2 00:00:02,760 --> 00:00:08,640 Previously we did find a generator through our class C which contains the architecture of the neural 3 00:00:08,640 --> 00:00:16,040 network inside the init function and the forward function to propagate the signal inside this architecture. 4 00:00:16,170 --> 00:00:21,660 And now that we have defined the class we were ready to create as many objects as we want there is as 5 00:00:21,660 --> 00:00:26,870 many generators as we want but we only need one and that's the one we'll create. 6 00:00:26,940 --> 00:00:28,460 And the Statoil. 7 00:00:28,530 --> 00:00:33,290 So to create an object of the class we need to choose a name for this object. 8 00:00:33,510 --> 00:00:40,350 And the name will choose is not g for as you might have guessed the neural network of the generator 9 00:00:41,090 --> 00:00:44,780 G and then to create a new object of the class. 10 00:00:44,840 --> 00:00:49,850 While there is nothing more simple you take your G class and then you add some parenthesis. 11 00:00:49,890 --> 00:00:52,170 Why do you only need to add some parenthesis. 12 00:00:52,290 --> 00:00:59,190 It's because in the arguments of the class we only inherited from the end module and we didn't put any 13 00:00:59,250 --> 00:00:59,890 argument. 14 00:00:59,940 --> 00:01:04,490 So basically there is no argument and therefore there is an argument in put here. 15 00:01:04,530 --> 00:01:06,410 Hence the only parenthesis. 16 00:01:06,540 --> 00:01:07,190 Perfect. 17 00:01:07,200 --> 00:01:12,910 And so in the flashiest of the flashes we got our generator neural network. 18 00:01:13,080 --> 00:01:14,250 Congratulations. 19 00:01:14,400 --> 00:01:21,390 And now as I said in the end of the previous Statoil we need to initialize the weights the proper way 20 00:01:21,510 --> 00:01:28,290 to respect the convention of the adversarial networks and to do this we have the weights in a function 21 00:01:28,290 --> 00:01:29,760 that can do that for us. 22 00:01:29,760 --> 00:01:32,450 So I'm quickly going to explain what it's going to do. 23 00:01:32,640 --> 00:01:36,780 As you can see we start with the class name variable. 24 00:01:36,780 --> 00:01:44,030 There is some kind of a research tool that will look for some names in the definition of the class so 25 00:01:44,040 --> 00:01:49,980 it will look for some names inside this class and the names it's going to look for are gone and Bajan 26 00:01:49,980 --> 00:01:54,360 on and since can be transposed to the contained can. 27 00:01:54,550 --> 00:01:57,220 Well it will find Canth transposed to the. 28 00:01:57,240 --> 00:02:06,090 And then it will initialize the weights to 0.00 and 0.02 for the convolution modules and then Same for 29 00:02:06,090 --> 00:02:06,550 Birgeneau. 30 00:02:06,570 --> 00:02:13,260 It's going to look for any name in the class that contains Bache norm which of course the budget norm 31 00:02:13,380 --> 00:02:16,110 to D2 budget normalized feature map. 32 00:02:16,180 --> 00:02:22,530 And on each of these layers related to the batched norms and inside each of these best layers it will 33 00:02:22,560 --> 00:02:26,140 initialize the weights to 1.0 0.02. 34 00:02:26,280 --> 00:02:34,140 And remember in each layer we also have some bias and all the bias at the batch on levels will be initialized 35 00:02:34,140 --> 00:02:34,790 to zero. 36 00:02:35,010 --> 00:02:41,160 So that's exactly what it's going to do and it's using this class name trick to look for the convolutions 37 00:02:41,370 --> 00:02:46,860 and the budget formalizations inside the class to initialize these ways the right way. 38 00:02:46,860 --> 00:02:48,560 All right so that's how it works. 39 00:02:48,600 --> 00:02:55,740 And now to apply this function we just need to take our generator neural network which is which we've 40 00:02:55,740 --> 00:03:05,070 just called Net G and then we added that then we're going to use the plie function to apply the weights 41 00:03:05,100 --> 00:03:05,760 in its function. 42 00:03:05,760 --> 00:03:11,510 So I'm just copying this and pasting it inside. 43 00:03:11,880 --> 00:03:16,290 All right and this will just apply the weight in function to our Najia object. 44 00:03:16,290 --> 00:03:19,360 That is the neural network of our generator. 45 00:03:19,380 --> 00:03:20,170 All right. 46 00:03:20,220 --> 00:03:21,590 So congratulations. 47 00:03:21,600 --> 00:03:26,250 We now have a generator a real generator neural network. 48 00:03:26,250 --> 00:03:31,830 So basically we are done with the first big step of this implementation of the deep convolutional Ganns 49 00:03:32,190 --> 00:03:35,940 which was all about defining and creating the generator. 50 00:03:35,940 --> 00:03:41,070 Now we're going to move on to the second big step of this implementation which will be about defining 51 00:03:41,250 --> 00:03:44,380 and creating this time the discriminator. 52 00:03:44,610 --> 00:03:46,840 So we'll do that in the next three to two year olds. 53 00:03:46,860 --> 00:03:51,870 We will start by defining the class then we'll define the forward function and then eventually we'll 54 00:03:51,870 --> 00:03:55,200 create our discriminator object. 55 00:03:55,200 --> 00:03:56,990 Until then enjoy computer vision. 5824