Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated: 1 00:00:00,420 --> 00:00:02,700 Hello and welcome to listen to Tauriel. 2 00:00:02,700 --> 00:00:08,220 All right so we have a series of transformations to make so that the input frame can be fed into the 3 00:00:08,220 --> 00:00:09,130 neural network. 4 00:00:09,270 --> 00:00:14,300 Let's do these four transformations in this tutorial and let's start immediately with the first one. 5 00:00:14,580 --> 00:00:20,910 So as we said in the previous Statoil the first one is to apply the transform transformation so that 6 00:00:21,120 --> 00:00:26,990 our input frame has a right format meaning the right time dominations and the right colors. 7 00:00:27,180 --> 00:00:34,710 So to apply this transformation we're going to introduce a new Voivode which will be frame T and that 8 00:00:34,710 --> 00:00:38,680 will respond to this future transform frame. 9 00:00:38,760 --> 00:00:44,190 But I'm calling it Frente in that frame because I want to keep the original frame because we will use 10 00:00:44,460 --> 00:00:47,700 the original frame and to put the rectangles inside. 11 00:00:47,960 --> 00:00:53,700 So Frente will be the transformed frame after applying the transform transformation and to apply this 12 00:00:53,700 --> 00:00:55,300 transform transformation. 13 00:00:55,500 --> 00:01:02,130 Well we simply need to take that transformation transform and as input as you might guess when put the 14 00:01:02,130 --> 00:01:11,340 frame frame then we must understand that this transform function returns two elements and we are only 15 00:01:11,340 --> 00:01:16,260 interested in the first element which is actually the transform frame the frame with the right format 16 00:01:16,710 --> 00:01:21,410 and therefore to get the first element returned by this function. 17 00:01:21,480 --> 00:01:27,960 We simply need to add some brackets here and get the index of the first element which is zero. 18 00:01:28,410 --> 00:01:32,140 All right so then we get this new transform frame. 19 00:01:32,160 --> 00:01:39,720 Now Frente has the right format that has the right dimensions and the right color values good first 20 00:01:39,720 --> 00:01:41,500 transformation done now. 21 00:01:41,520 --> 00:01:43,080 Second transformation. 22 00:01:43,170 --> 00:01:45,810 Remember Frente is an umpire. 23 00:01:45,930 --> 00:01:50,210 We applied the first transformation but it still returns a number by Array. 24 00:01:50,340 --> 00:01:56,690 And so the second transformation to make now is to convert this number by Array into the torch tensor. 25 00:01:56,740 --> 00:02:04,140 I remind the torch tensor is a much more advanced matrix of a single type than an array. 26 00:02:04,260 --> 00:02:08,310 It is very useful for neural networks but it will not be useful enough. 27 00:02:08,310 --> 00:02:12,040 You will see we will have still two transformations to make. 28 00:02:12,090 --> 00:02:17,240 So let's do this second transformation converting the non-bio right into a torch sensor. 29 00:02:17,280 --> 00:02:24,780 So since we're getting closer and closer to the final input that will be accepted by our as is the neural 30 00:02:24,780 --> 00:02:25,480 network. 31 00:02:25,650 --> 00:02:32,340 I'm going to start to call this input X because then I will just override the x variable with the same 32 00:02:32,340 --> 00:02:32,870 name. 33 00:02:33,060 --> 00:02:39,520 So X will be this tortured answer that will just be converted from the non-firing. 34 00:02:39,810 --> 00:02:45,150 So now it's very simple to convert a non binary into a tortured answer. 35 00:02:45,150 --> 00:02:53,070 We just need to take our torched library and then a simple function very intuitive to remember which 36 00:02:53,070 --> 00:02:55,840 is from none. 37 00:02:56,130 --> 00:03:02,970 All right from run by the function that converts numbers into torsion answers and therefore very obviously 38 00:03:03,090 --> 00:03:07,240 what we have to input here in this function is of course our number. 39 00:03:07,250 --> 00:03:07,650 Right. 40 00:03:07,650 --> 00:03:10,870 That is Frente Okay perfect. 41 00:03:10,950 --> 00:03:16,250 But now there is another slight thing to do we could call it a sub transformation. 42 00:03:16,250 --> 00:03:23,400 This wasn't a transformation that I mentioned because it's a small thing but anyway this small transformation 43 00:03:23,430 --> 00:03:31,340 is to convert the sequence red blue green into green red blue because right now the order of the indexes 44 00:03:31,340 --> 00:03:34,560 or the color for our image is red blue green. 45 00:03:34,680 --> 00:03:40,890 But the neural network says he was trained under the convention green red blue. 46 00:03:40,920 --> 00:03:46,260 So we just need to inverse the order but that's a specific thing to the neural network and therefore 47 00:03:46,530 --> 00:03:52,320 this is not the general transformations that we have to make each time remember that the transformations 48 00:03:52,320 --> 00:03:58,920 that we're making except the one we're about to make right now is the general process before finding 49 00:03:58,970 --> 00:04:04,710 and then put into a new network implemented with torche that's very important to remember but by doing 50 00:04:04,710 --> 00:04:07,330 it two or three times it will be very easy for you. 51 00:04:07,380 --> 00:04:12,360 So let's do this small and quick transformation that is a permanent nation of colors. 52 00:04:12,540 --> 00:04:18,510 So we're going to add a dot here and then we're going to use the permute function and we want to go 53 00:04:18,510 --> 00:04:21,870 from red blue green to green red blue. 54 00:04:21,930 --> 00:04:29,550 So since Green is the last index number two we always to hear then since Red is the first one we put 55 00:04:29,620 --> 00:04:36,000 a zero and since Blue is the second index that is one we put one right. 56 00:04:36,000 --> 00:04:39,220 We want to go from red blue green to green red blue. 57 00:04:39,220 --> 00:04:43,030 So we want to go from 0 1 to 2 2 0 1. 58 00:04:43,170 --> 00:04:44,040 Perfect. 59 00:04:44,040 --> 00:04:50,850 So now we have the right torch sensor format with the right order of colors. 60 00:04:50,850 --> 00:04:51,570 Perfect. 61 00:04:51,570 --> 00:04:56,220 Now next transformation this is the third transformation out of the four. 62 00:04:56,310 --> 00:05:01,450 But actually we're going to make the last two transformations in the same one line of code. 63 00:05:01,470 --> 00:05:07,810 So the third transformation is to add this fake dimension corresponding to the batch. 64 00:05:07,830 --> 00:05:15,300 And the reason for doing this is that the neural network cannot actually accept single inputs like a 65 00:05:15,300 --> 00:05:18,810 single input vector or a single input image. 66 00:05:18,900 --> 00:05:22,540 It only accepts them in to some batches. 67 00:05:22,560 --> 00:05:27,040 So basically the neural network only except batches of inputs. 68 00:05:27,180 --> 00:05:33,390 And that's why now we have to create a structure with the first time engine correspond to the batch 69 00:05:33,750 --> 00:05:37,480 and then the other dimensions corresponding to the input. 70 00:05:37,680 --> 00:05:40,440 That's very very important in PI torch. 71 00:05:40,470 --> 00:05:44,500 You will always see that we do it with the squeeze function. 72 00:05:44,580 --> 00:05:50,490 So each time you see the squeeze function it's definitely just before feeding the neural network with 73 00:05:50,490 --> 00:05:54,820 the input we use the N squeeze function to create that thing domination of the batch. 74 00:05:54,840 --> 00:06:00,940 So good thing to keep in mind if you're going to use torture in the future which I highly recommend. 75 00:06:01,110 --> 00:06:03,700 But let's do this third transformation. 76 00:06:03,830 --> 00:06:07,440 The Squeeze and the good news is that it's very simple. 77 00:06:07,440 --> 00:06:12,350 We take our input image which is now a torch sensor. 78 00:06:12,450 --> 00:06:20,340 Thanks to the previous transformation we are added and then we use the squeeze function. 79 00:06:20,910 --> 00:06:28,950 And now this squeeze function takes one argument which is exactly the index of the dimension of the 80 00:06:28,950 --> 00:06:33,290 batch and the batch should always be the first time engine. 81 00:06:33,480 --> 00:06:39,700 So it you always have the first index and therefore since indexes and bytes and start at zero. 82 00:06:39,870 --> 00:06:42,700 Well we put a zero that zero. 83 00:06:42,700 --> 00:06:48,780 Here is the Index of this first diamond and corresponding to the bet that we're adding to our structure 84 00:06:48,900 --> 00:06:50,300 of input images. 85 00:06:51,460 --> 00:06:53,810 All right so that's the third transformation that's done. 86 00:06:53,860 --> 00:06:54,760 Perfect. 87 00:06:54,850 --> 00:06:58,350 And now we're going to do the last transformation in that same line. 88 00:06:58,600 --> 00:07:00,980 And that last transformation. 89 00:07:01,030 --> 00:07:06,970 The final one and then we'll be ready to feed the new one that work with the input image is to convert 90 00:07:07,240 --> 00:07:16,180 this batch of torture and sort of input into a torch viable I remind a torch viable is a highly advanced 91 00:07:16,180 --> 00:07:20,080 variable that contains both a tensor and a gradient. 92 00:07:20,080 --> 00:07:25,810 This torch viable will become an element of the dynamic graph which will compute very efficiently the 93 00:07:25,810 --> 00:07:30,480 gradients of any composition functions during backward propagation. 94 00:07:30,490 --> 00:07:31,340 So there we go. 95 00:07:31,360 --> 00:07:33,150 That's the final transformation. 96 00:07:33,190 --> 00:07:34,880 And that is very easy to do. 97 00:07:34,930 --> 00:07:41,880 We simply need to take the variable class like that. 98 00:07:41,920 --> 00:07:48,630 So this is the variable class which will create an object which will be this towards variable. 99 00:07:48,760 --> 00:07:54,730 And therefore since we're creating a new object we need to override the previous variable X because 100 00:07:54,730 --> 00:08:01,720 I'm going to call it X again and I'm going to add here X equals variable x and zero. 101 00:08:02,080 --> 00:08:06,870 And that's exactly an object which is the torch viable. 102 00:08:06,890 --> 00:08:10,090 So now purrfect are four transformations are done. 103 00:08:10,150 --> 00:08:17,590 So congratulations you are ready to feed the is the neural network with the input images that are now 104 00:08:17,770 --> 00:08:18,940 towards variables. 105 00:08:19,180 --> 00:08:20,440 So that's wonderful. 106 00:08:20,440 --> 00:08:21,900 We're done with this material. 107 00:08:21,910 --> 00:08:24,730 We did very well our four transformations. 108 00:08:24,730 --> 00:08:31,600 So in the next time we will feed this torch very well to the neural network SSD which is already pre-trained 109 00:08:31,630 --> 00:08:37,870 because you know it's a neural network that was pre-trained to detect between 30 to 40 objects including 110 00:08:37,870 --> 00:08:42,500 planes horses dogs cars boats a lot more. 111 00:08:42,580 --> 00:08:44,820 So we're not going to do the whole training again. 112 00:08:44,830 --> 00:08:45,930 That's the beauty of it. 113 00:08:45,940 --> 00:08:47,860 We have a pre-trained model. 114 00:08:47,880 --> 00:08:51,730 We're going to feed the input image to this pre-trained model and we will get the output. 115 00:08:51,790 --> 00:08:54,470 That is the prediction detection. 116 00:08:54,970 --> 00:08:58,720 So let's do this in the next little while and until then enjoy computer vision. 12714