1 00:00:00,330 --> 00:00:05,910 Hi and welcome to this part of the course where we actually get our feet wet and start using some code 2 00:00:05,910 --> 00:00:10,370 and building our first convolutional neural network using Python. 3 00:00:11,280 --> 00:00:16,230 So now let's dive into the code into Google Cloud and get started with this lesson. 4 00:00:17,400 --> 00:00:24,990 OK, so you're here on Google Cloud and I assume you've unzipped all of the course coded into your Google 5 00:00:24,990 --> 00:00:32,640 Drive, and you can place the code at the course code called Modern Computer Vision to Folder and you'll 6 00:00:32,640 --> 00:00:34,070 call up in the books directory. 7 00:00:34,110 --> 00:00:36,940 It's fine if you leave it outside, you can put it any way you want. 8 00:00:36,960 --> 00:00:42,870 I just have it here because it's good to keep the stuff organized, so we have folders here. 9 00:00:43,380 --> 00:00:46,830 If you've done the open to be part of the course, you would have seen all of the course. 10 00:00:46,830 --> 00:00:50,910 Could a full open CV right here, the all the fortysomething notebooks. 11 00:00:51,540 --> 00:00:55,080 Now we're going to start the deep learning part of the course. 12 00:00:55,590 --> 00:01:01,530 So let's dive into the deep learning part of the diplomatic folder and you'll see all of the course. 13 00:01:01,530 --> 00:01:02,000 Good here. 14 00:01:02,010 --> 00:01:07,950 There's a lot of notebooks here, a lot of projects, a lot of invaluable lessons because I'm a computer 15 00:01:07,950 --> 00:01:14,310 vision engineer practitioner and researcher, and I routinely routinely look up my own notebooks because 16 00:01:14,310 --> 00:01:16,620 there's just so much knowledge and you tend to forget things. 17 00:01:17,070 --> 00:01:20,640 So it's a very good resource for just having all of these notebooks here. 18 00:01:21,300 --> 00:01:27,090 So the first one we're going to do is to pay to watch CNN tutorial using the amnesty dataset. 19 00:01:27,540 --> 00:01:33,360 Now the amnesty, the said, is a set of a very famous computer vision data set of handwritten digits 20 00:01:33,780 --> 00:01:35,760 that was compiled by USPS. 21 00:01:36,300 --> 00:01:37,590 Maybe in the 90s, I believe. 22 00:01:38,040 --> 00:01:43,800 And it's been the benchmark for early computer vision algorithms and then deleted scenes and deep of 23 00:01:43,800 --> 00:01:45,840 the deep different stuff over the years. 24 00:01:46,140 --> 00:01:50,070 So it's a really good baseline dataset to use to experiment on. 25 00:01:50,790 --> 00:01:52,540 So let's open this notebook out. 26 00:01:52,560 --> 00:01:55,410 I already have it up in here because I was just reading some stuff before. 27 00:01:56,190 --> 00:02:02,660 So let's take a tour of the notebook this you can see we have the definition of what we're doing here, 28 00:02:02,680 --> 00:02:09,780 creating a simple CNN amnesty to set, which is handwritten digits and a list of 10 different things 29 00:02:09,780 --> 00:02:13,080 here that we're going to do going to be doing in this notebook. 30 00:02:13,680 --> 00:02:20,670 So a lot of my notebooks in this section in this part of Texas have numbers one, two three, whatever, 31 00:02:21,510 --> 00:02:22,620 and that's how we teach. 32 00:02:23,190 --> 00:02:29,040 I teach in blocks and each block is meant to be pretty, pretty simple, small lesson. 33 00:02:29,520 --> 00:02:33,540 And it's I think it's a good way to split up the lessons here. 34 00:02:33,990 --> 00:02:35,430 So let's get started. 35 00:02:35,520 --> 00:02:39,660 So firstly, we want to build a CNN using by touch. 36 00:02:39,810 --> 00:02:40,740 What do we need to do? 37 00:02:41,130 --> 00:02:43,560 Well, we need to import these libraries. 38 00:02:44,190 --> 00:02:48,600 So the first library you're going to import is PyTorch and to import PyTorch. 39 00:02:48,930 --> 00:02:54,210 It's actually just a short torch when we when Facebook made this thing, called the torch. 40 00:02:54,720 --> 00:02:56,700 So we just import torch. 41 00:02:56,760 --> 00:02:58,830 We don't actually have to put the word play torch. 42 00:02:59,190 --> 00:03:03,210 It's a bit confusing sometimes, because that's a foolish offensive to carry out such a scare us. 43 00:03:03,750 --> 00:03:09,480 But a lot of libraries do have different names, depending on why maybe there was existing play torch 44 00:03:09,480 --> 00:03:09,930 for those. 45 00:03:10,470 --> 00:03:12,360 But we have a torch here. 46 00:03:13,260 --> 00:03:20,250 Next, we import something called torch vision, and torch vision is a set of accompanying libraries 47 00:03:20,250 --> 00:03:23,580 that go together with the Torch Library or the Torch Library. 48 00:03:23,940 --> 00:03:29,280 I'm going to use PI torch and torch interchangeably and most likely will simply torch because it's just 49 00:03:29,940 --> 00:03:31,080 easier to say, to be fair. 50 00:03:31,590 --> 00:03:34,830 So let's take a look at this here. 51 00:03:35,340 --> 00:03:36,420 So what? 52 00:03:36,480 --> 00:03:37,860 Why do we need torch vision? 53 00:03:38,370 --> 00:03:45,630 Well, Torch Mission has a lot of data sets and other useful utility functions that we use with the 54 00:03:45,630 --> 00:03:48,800 Torch Library to make things much easier. 55 00:03:48,810 --> 00:03:49,860 So you'll get started. 56 00:03:49,860 --> 00:03:50,840 Don't worry about this yet. 57 00:03:50,850 --> 00:03:53,790 You'll get started and you'll see what all these libraries do. 58 00:03:54,120 --> 00:03:57,090 So don't don't get consumed too much yet. 59 00:03:57,690 --> 00:03:59,730 It's all going to make sense shortly. 60 00:04:00,480 --> 00:04:05,680 So you see, we import two things we import what vision and then we import the specific function here 61 00:04:05,700 --> 00:04:06,530 transforms. 62 00:04:07,350 --> 00:04:10,950 And we want to be important to watch vision that transforms and we call it transforms. 63 00:04:10,950 --> 00:04:14,130 And it could transform the very useful functions. 64 00:04:14,130 --> 00:04:17,250 And I'll demonstrate why later on. 65 00:04:18,030 --> 00:04:22,410 Next, we import some torch libraries here that we are going to use to build CNN. 66 00:04:22,890 --> 00:04:27,370 So we import torchlight optim as optim as well as entertaining. 67 00:04:27,630 --> 00:04:33,480 And what we used as command in Python is just so that we don't have to write this continuously. 68 00:04:33,480 --> 00:04:33,900 No code. 69 00:04:34,380 --> 00:04:35,750 Makes it easier to read. 70 00:04:36,000 --> 00:04:41,280 So it's a good thing to import a function like this towards the OPTIM as often. 71 00:04:42,060 --> 00:04:46,770 But this is really that object technically right up here importing not the function. 72 00:04:47,520 --> 00:04:52,050 And then lastly, is an important line that you constantly see in all the Python Watch lessons. 73 00:04:52,620 --> 00:04:53,640 I'll be using CUDA. 74 00:04:53,790 --> 00:04:54,800 So what is CUDA? 75 00:04:55,110 --> 00:04:59,730 Well, I'll probably put up a lesson about CUDA afterward, but. 76 00:05:00,250 --> 00:05:08,050 Is the video deep learning both Nvidia duping you accelerating library, that deep learning libraries 77 00:05:08,050 --> 00:05:11,740 used to do the calculations in the back end speed of these things on GPUs? 78 00:05:12,340 --> 00:05:19,960 So if CUDA is available, PyTorch will use CUDA and it'll become vastly faster and more efficient. 79 00:05:20,380 --> 00:05:26,020 However, if CUDA is unavailable, then you're going to be using the CPU instead, which is a lot slower. 80 00:05:26,230 --> 00:05:27,590 However, it still does work. 81 00:05:27,640 --> 00:05:34,510 So to get the GPU on CoLab, this notebook should open automatically with the GPU. 82 00:05:34,990 --> 00:05:40,960 However, if you go to edit notebook settings, you can see you can get the GPU, CPU and none. 83 00:05:41,380 --> 00:05:45,440 That means you're using strictly the CPU can even increase your RAM. 84 00:05:45,460 --> 00:05:50,230 Most of our listeners don't need to increase to higher, so that's fine. 85 00:05:50,230 --> 00:05:56,410 You can leave it as is, but if you're if your notebook is sitting on none, change it to GPU and then 86 00:05:57,040 --> 00:05:58,140 won't change the GPU. 87 00:05:58,240 --> 00:06:01,990 The notebook will restart so we can lose everything for that point. 88 00:06:02,290 --> 00:06:09,970 As in all the things, he said, the notebook will still be there and for no change in the GPU and run 89 00:06:09,970 --> 00:06:13,000 this block of code, which I've already run, says GPU available. 90 00:06:13,720 --> 00:06:18,130 And it takes about five or six seconds to import these packages, and you only need to do this once 91 00:06:18,310 --> 00:06:19,030 in your notebook. 92 00:06:19,450 --> 00:06:20,620 You don't need to do this again. 93 00:06:21,040 --> 00:06:29,440 So the torch that CUDA is available is a function that tells us a Boolean Boolean function sends a bullet 94 00:06:29,440 --> 00:06:34,060 argument that tells us what the CUDA is available and what that means. 95 00:06:34,060 --> 00:06:37,240 You going to be using the GPU if it is or CPU, if it's not. 96 00:06:37,780 --> 00:06:43,280 And we have this device variable we use to call it, but occasionally went by to its functions that 97 00:06:43,280 --> 00:06:45,730 would be need to send things to the CPU or CPU. 98 00:06:46,360 --> 00:06:47,960 So don't worry too much about this yet. 99 00:06:47,980 --> 00:06:52,150 This is just setting sphere to what we're using, so we want to be using the GPU. 100 00:06:52,540 --> 00:06:59,450 So if cooler is available, it will automatically set device equal to CUDA and any any function we use 101 00:06:59,450 --> 00:07:02,650 and pay too much to has the device as an input parameter. 102 00:07:02,660 --> 00:07:04,090 It'll know what it wants to. 103 00:07:04,220 --> 00:07:12,130 We want to use GPU, so let's stop there for now and we'll move on to the transformer lesson. 104 00:07:12,430 --> 00:07:13,900 So I'll see you in the next section.