Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated: 1 00:00:00,420 --> 00:00:05,720 Hello and welcome to the second big part of this implementation training the brains. 2 00:00:05,880 --> 00:00:11,520 So we have just smashed part one creating the brains and now we're going to tackle part to training 3 00:00:11,610 --> 00:00:12,480 the brains. 4 00:00:12,510 --> 00:00:16,580 So this training implementation will be broken down into two big steps. 5 00:00:16,590 --> 00:00:22,230 First step will be to update the weight of the neural network of the discriminator and then the second 6 00:00:22,380 --> 00:00:26,370 step will be to update the weight of the new one that work of the generator. 7 00:00:26,460 --> 00:00:30,460 And then if we want to break down this process into some more steps. 8 00:00:30,570 --> 00:00:33,520 Well that's how it goes in the first big step. 9 00:00:33,540 --> 00:00:39,840 Updating the weights of the discriminator we will want to train the discriminator to see to understand 10 00:00:40,110 --> 00:00:41,590 what's real and what's fake. 11 00:00:41,790 --> 00:00:48,030 So we will first train it by giving it a real image and we'll set the target to one because one means 12 00:00:48,210 --> 00:00:54,570 that the image is accepted and then we'll do another training by giving it this time a fake image and 13 00:00:54,570 --> 00:00:59,170 setting the standard target to zero because zero means that the image is not accepted. 14 00:00:59,400 --> 00:01:04,120 So this first big step can be broken down into two steps. 15 00:01:04,170 --> 00:01:09,960 First training the discriminator with a real image and then training the discriminator with a fake image 16 00:01:10,380 --> 00:01:14,590 and this fake image will be of course a fake image created by the generator. 17 00:01:14,970 --> 00:01:22,270 And then in the second step of the training that is about the weight of the neural network of the generator. 18 00:01:22,440 --> 00:01:28,710 Well we'll take the fake image again which will be in one step of the loop then we'll feed this fake 19 00:01:28,710 --> 00:01:35,130 image into the discriminator to get the output which will be a value between 0 and 1 the discriminating 20 00:01:35,130 --> 00:01:35,750 value. 21 00:01:35,970 --> 00:01:38,770 But then we'll set a new target to 1. 22 00:01:38,820 --> 00:01:44,520 The target will be equal to 1 always and will compute the computerless between the output of the discriminator 23 00:01:44,640 --> 00:01:46,250 the value between 0 and 1. 24 00:01:46,290 --> 00:01:49,890 And this target always equal to 1 then be careful. 25 00:01:49,890 --> 00:01:55,920 That's the key thing to understand will back propagate this error not back inside the discriminator 26 00:01:56,100 --> 00:01:58,320 but back inside the generator. 27 00:01:58,320 --> 00:02:03,450 That's the key thing to understand the error is the error between the prediction of the discriminator 28 00:02:03,720 --> 00:02:09,750 and the target equal to one but will back propagate this error back inside the neural network of the 29 00:02:09,750 --> 00:02:10,580 generator. 30 00:02:10,890 --> 00:02:16,500 And then once we do that will applies to kesa great in the sense to date the weight of the neural network 31 00:02:16,740 --> 00:02:17,820 of the generator. 32 00:02:18,120 --> 00:02:18,700 All right. 33 00:02:18,750 --> 00:02:25,040 So we have given a teaser of what's going to happen in the second phase of training the brain. 34 00:02:25,140 --> 00:02:29,110 So let's tackle the second phase and let's start right now. 35 00:02:29,460 --> 00:02:29,810 All right. 36 00:02:29,820 --> 00:02:36,030 So we have to start by getting a criterion that will measure the error of prediction that will be of 37 00:02:36,030 --> 00:02:41,640 value between 0 and 1 because that will be the prediction of the discriminator the discriminating number 38 00:02:41,640 --> 00:02:49,910 between 0 and 1 and a ground truth that will only be 0 or 1 0 means fake and 1 means real. 39 00:02:49,950 --> 00:02:50,860 So let's do this. 40 00:02:50,940 --> 00:02:52,340 That's our first step. 41 00:02:52,350 --> 00:03:02,500 We create a criterion object criterion object that will be an object of the b c the last. 42 00:03:02,520 --> 00:03:09,390 So if we go to the PI torch documentation we can see that the B C class is actually given by this formula. 43 00:03:09,630 --> 00:03:17,010 And so this is a special kind of less perfect to train at this adversarial networks and busy means binary 44 00:03:17,010 --> 00:03:18,120 Krus entropy. 45 00:03:18,360 --> 00:03:21,350 So that's the last we'll be using for our Just again. 46 00:03:21,360 --> 00:03:24,630 But we see that it is also used for our encounters. 47 00:03:24,810 --> 00:03:31,140 But the most important thing is that the target should be numbers between 0 and 1 and that will be the 48 00:03:31,140 --> 00:03:32,580 case for TCN. 49 00:03:32,610 --> 00:03:38,550 Because the targets will either be zero when we want to try to discriminate or to recognize a fake image 50 00:03:38,970 --> 00:03:39,900 or one. 51 00:03:39,900 --> 00:03:43,420 When we train the discriminator to recognize a real image. 52 00:03:43,480 --> 00:03:46,940 So basically us and then we just need some parenthesis. 53 00:03:46,950 --> 00:03:48,510 All right so perfect. 54 00:03:48,510 --> 00:03:55,060 We have our criterion and now we need to optimize one optimizer for the generator and one optimized 55 00:03:55,060 --> 00:03:56,650 for the discriminator. 56 00:03:56,670 --> 00:04:01,590 So we'll start with the optimizer of the discriminator. 57 00:04:01,860 --> 00:04:04,190 And this time we're not going to get it from. 58 00:04:04,260 --> 00:04:10,420 And then the end module but if we have a look again at our libraries that we imported. 59 00:04:10,500 --> 00:04:13,440 Well as you can see here we imported towards start up. 60 00:04:13,560 --> 00:04:16,850 And we gave it the shortcut name up to him to simplify. 61 00:04:16,980 --> 00:04:21,700 But that's from which we will import our optimizer object. 62 00:04:21,720 --> 00:04:31,380 So here we go optimize Riddhi equals Optum dot and we'll get the atom optimizer which is a highly advanced 63 00:04:31,440 --> 00:04:34,090 optimizer for stochastic great in the sense. 64 00:04:34,250 --> 00:04:34,970 Okay. 65 00:04:35,220 --> 00:04:40,770 And now this time we have to put some arguments the first argument we need to input are the parameters 66 00:04:40,860 --> 00:04:47,310 of our neural network to one of the discriminator and to get them we're going to take our neural network 67 00:04:47,400 --> 00:04:50,990 objects of the discriminator that we called Net D. 68 00:04:51,090 --> 00:04:52,900 So net D. 69 00:04:53,190 --> 00:05:01,320 And then that and then we get the parameters this way with some parenthesis then we specify a learning 70 00:05:01,320 --> 00:05:10,830 rate and the argument for that is are and will show the value of our point 0 2 and then the third argument 71 00:05:10,860 --> 00:05:16,710 we need to specify the beta's parameters and we're going to improve them this way the name of the argument 72 00:05:16,770 --> 00:05:18,150 is better. 73 00:05:18,330 --> 00:05:22,310 And it's actually a couple of two values that we have to choose. 74 00:05:22,500 --> 00:05:29,310 And if we go to the pie torch documentation again well we see that for this Adam optimizer that these 75 00:05:29,320 --> 00:05:35,340 bettors are actually coefficients used for computing running averages of gradient and square. 76 00:05:35,370 --> 00:05:42,930 So we're just going to pick two values now again coming from experimentation and we'll choose 0.5 and 77 00:05:43,440 --> 00:05:45,300 point nine nine. 78 00:05:45,510 --> 00:05:46,180 Great. 79 00:05:46,230 --> 00:05:47,160 Perfect. 80 00:05:47,160 --> 00:05:50,590 We have the optimizer of the discriminator. 81 00:05:50,700 --> 00:05:54,430 And now let's get the optimizer of the generator. 82 00:05:54,630 --> 00:05:59,670 So since we're going to have the same parameters the same learning rate and the same betas. 83 00:05:59,820 --> 00:06:03,270 We simply need to change the name of the optimizer. 84 00:06:03,420 --> 00:06:10,690 And of course we're going to call the optimize of the generator optimizer G which will also be an atom 85 00:06:10,770 --> 00:06:16,380 optimizer with the same parameters except of course for the parameters of the neural network. 86 00:06:16,380 --> 00:06:21,630 So here we just need to replace Nat d by Najee and perfect. 87 00:06:21,630 --> 00:06:29,040 Now we have zero Pretorian attached to the PC plus the optimizer of the discriminator and the optimizer 88 00:06:29,040 --> 00:06:30,230 of the generator. 89 00:06:30,510 --> 00:06:36,840 And that's good news because basically we're ready to start the big loop of the training that is the 90 00:06:36,840 --> 00:06:39,120 loop over the different epochs. 91 00:06:39,270 --> 00:06:49,410 So we're going to have 25 EPOC and therefore this loop is simply going to be for Epoque in range 25 92 00:06:50,080 --> 00:06:50,720 and in. 93 00:06:50,900 --> 00:06:55,340 And this way the epochs will go from zero to 24. 94 00:06:55,470 --> 00:06:57,070 We'll have 25 epix. 95 00:06:57,150 --> 00:07:02,580 You're welcome to try more e-books if you want to try to have some even better images. 96 00:07:02,580 --> 00:07:07,540 But I can tell you that already with 25 bucks we'll get some great images. 97 00:07:07,860 --> 00:07:14,970 And also it's important to understand is that in each of these 25 bucks we'll go through all the images 98 00:07:15,060 --> 00:07:19,620 of the dataset we'll go through all the images of the dataset twenty five times. 99 00:07:19,950 --> 00:07:26,130 All right so now we're directly going to start with a new for loop that if you've listened to the last 100 00:07:26,220 --> 00:07:32,150 sentence I've just said it's going to be the nucleus we'll go through all the images of the dataset. 101 00:07:32,220 --> 00:07:39,780 Therefore the second for loop is going to be for I which will just be the index of the loop and then 102 00:07:40,320 --> 00:07:44,020 data that will contain a million batch of images. 103 00:07:44,130 --> 00:07:48,450 We will go through all the images of the dataset mini batch by mini bitch. 104 00:07:48,540 --> 00:07:49,720 That is what it means. 105 00:07:49,890 --> 00:07:52,890 And data will be one of these mini batches. 106 00:07:53,100 --> 00:07:59,580 So we're breaking the data the whole data set of all the images in too many batches and data will be 107 00:07:59,760 --> 00:08:03,020 each of these different mini batches. 108 00:08:03,090 --> 00:08:08,970 And so now to get these different mini batches we're going to use data loader and we're going to use 109 00:08:08,970 --> 00:08:15,760 the enemy rate function to get these separate mini batches of the dataset. 110 00:08:16,170 --> 00:08:23,430 And so inside and emirate's we have to put our data loader which will get us the mini batches but then 111 00:08:23,430 --> 00:08:29,100 we also have to specify where the index of the loop is going to start from that is where I is going 112 00:08:29,100 --> 00:08:29,980 to start from. 113 00:08:30,150 --> 00:08:31,650 And I will start from zero. 114 00:08:31,740 --> 00:08:35,100 So I'm just adding 0 here and there we go. 115 00:08:35,130 --> 00:08:38,880 We are ready to start the big two steps of the training. 116 00:08:38,880 --> 00:08:41,920 So we'll do these big two steps in two separate tutorials. 117 00:08:41,940 --> 00:08:46,560 The next one will be about updating the weights of the neural network the discriminator. 118 00:08:46,560 --> 00:08:51,300 Then the next tutorial after that will be about the second big step about the update of the weight of 119 00:08:51,300 --> 00:08:58,110 the new work of the generator and then we'll have the last final exciting tutorial to print the classes 120 00:08:58,260 --> 00:09:02,970 save the real images save the fake images and watch the final results. 121 00:09:02,970 --> 00:09:04,390 So be ready for all this. 122 00:09:04,410 --> 00:09:06,310 And until then enjoy computer vision. 13242