1 00:00:00,570 --> 00:00:07,710 And welcome back that, of course, in this section will talk about something or form a project that 2 00:00:07,710 --> 00:00:10,530 actually is quite popular and in demand. 3 00:00:10,950 --> 00:00:16,830 And that's how do we actually deploy our computer vision models and one of the many ways to deploy, 4 00:00:16,830 --> 00:00:21,960 actually, I'm going to record a whole section on production deployment later on. 5 00:00:22,560 --> 00:00:29,760 However, the easiest way right now is to create a rest API that allows you to send images to that API 6 00:00:29,760 --> 00:00:36,930 access point, and that API then returns its class or whatever you wanted to do with that image. 7 00:00:37,200 --> 00:00:42,200 You can actually turn the entire image if you were doing something like low light enhancement or neural 8 00:00:42,200 --> 00:00:47,640 old style transfer, or we can just give you an object detector overlay, or it can give you the class 9 00:00:47,640 --> 00:00:48,300 of the image. 10 00:00:48,870 --> 00:00:56,400 So let's take a look at how we can use Flask, which is a Python micro web framework for micro sites 11 00:00:57,000 --> 00:01:00,480 that we we're going to use that to create, firstly, an API. 12 00:01:00,900 --> 00:01:06,600 So let's create the API civil opening notebook 73 here, and we'll begin. 13 00:01:06,750 --> 00:01:12,750 So before I begin, actually, I just wanted to mention a bit of what flask of Flask is a web framework 14 00:01:13,080 --> 00:01:14,370 development library. 15 00:01:14,730 --> 00:01:16,170 It's a very, very good one. 16 00:01:16,170 --> 00:01:18,540 I strongly preferred the jungle. 17 00:01:18,600 --> 00:01:23,880 Even though Django is better, Django makes things a lot more complicated than it needs to be, in my 18 00:01:23,880 --> 00:01:24,300 opinion. 19 00:01:25,200 --> 00:01:27,930 And Flask is actually quite easy to get up and running. 20 00:01:28,380 --> 00:01:35,040 It's quite easy to understand how it's how it's structured, how the templates work, how everything 21 00:01:35,040 --> 00:01:36,090 kind of goes together. 22 00:01:36,510 --> 00:01:38,930 And it actually has a lot of advanced features. 23 00:01:38,940 --> 00:01:46,950 So Flask is pretty much capable of almost everything Django is, except maybe some more advanced functionality 24 00:01:46,950 --> 00:01:48,180 and management tools. 25 00:01:48,300 --> 00:01:50,840 So that's it for Flask. 26 00:01:50,850 --> 00:01:58,200 And now we're also going to need to install and grow and grow up allows us to access those local ports 27 00:01:58,200 --> 00:02:01,050 here on this call out machine on the internet. 28 00:02:01,260 --> 00:02:04,890 So which is what we're going to be doing, exposing our API? 29 00:02:05,100 --> 00:02:06,300 So let's get started. 30 00:02:06,300 --> 00:02:13,320 So clearly does my repo here with the flask up, then we install and grok here pip, install, flash 31 00:02:13,320 --> 00:02:13,920 and grok. 32 00:02:14,670 --> 00:02:21,330 Next, we need to actually download the grok system for Linux, so we don't want that file here and 33 00:02:21,330 --> 00:02:22,540 then we need to install it. 34 00:02:22,560 --> 00:02:30,930 This takes about 10 11 seconds to install and know what I need you to do is actually put a link in the 35 00:02:30,930 --> 00:02:33,680 notebook for you guys to sign up for and grok accounts. 36 00:02:33,690 --> 00:02:35,910 Can you see which gets better? 37 00:02:35,910 --> 00:02:37,410 But Google is quite quick. 38 00:02:37,920 --> 00:02:45,120 And then navigate to getting to your token and you just put exclamation mark and grok or token. 39 00:02:45,510 --> 00:02:47,190 And this saves your configuration. 40 00:02:47,400 --> 00:02:50,010 You took into a configuration file here. 41 00:02:50,430 --> 00:02:56,970 I've already done it, which is why I've run this number before and I've had my could my token here, 42 00:02:56,970 --> 00:02:58,470 so you guys can't access it. 43 00:02:59,100 --> 00:03:02,110 So in the meantime, let's just put a link here. 44 00:03:02,130 --> 00:03:04,740 This is the signup link. 45 00:03:07,420 --> 00:03:12,610 So once we do that now, we never get back into the directory and we're ready to get started here. 46 00:03:13,030 --> 00:03:17,650 So Flask is quite easy to use, so all you have to do is just import your libraries that you're going 47 00:03:17,650 --> 00:03:18,310 to be using. 48 00:03:18,310 --> 00:03:23,110 So we're using are you adjusting the torch vision for the model because we're going to be loading a 49 00:03:23,110 --> 00:03:25,630 pre-trained image that pie PyTorch model? 50 00:03:26,320 --> 00:03:30,160 We import pills and image processing processing tools. 51 00:03:30,160 --> 00:03:33,520 We import different things on flask to request that you solidify. 52 00:03:34,210 --> 00:03:37,330 And we also use flash and drop to run with ink. 53 00:03:37,870 --> 00:03:43,390 So we declared up here optical flask on the school underscore name, underscore underscore. 54 00:03:44,140 --> 00:03:47,740 Then we have run away and grow the app that we named him. 55 00:03:48,790 --> 00:03:53,980 Next, we need to load our pre-trained model, which is what we do in these three lines here, and we 56 00:03:53,980 --> 00:03:55,750 set the model to evaluation mode. 57 00:03:55,780 --> 00:04:02,260 And now we have a helper function that allows us to take an image that we are uploading from an outside 58 00:04:02,740 --> 00:04:11,350 the client and basically transform that that image to running these pre-processing functions so that 59 00:04:11,350 --> 00:04:16,270 then we can now send it to our we use it with another get prediction function. 60 00:04:16,780 --> 00:04:22,660 So that takes the image that we get in bytes, transforms the image into a tensor and then passes that 61 00:04:22,660 --> 00:04:24,010 tensor to the model. 62 00:04:24,430 --> 00:04:27,070 Then and then it returns to Class Hill. 63 00:04:27,460 --> 00:04:29,290 So that's quite simple. 64 00:04:29,410 --> 00:04:35,740 And now all you need to do in Flask Flask basically runs is to arrest API takes requests. 65 00:04:36,130 --> 00:04:42,110 So if we send a post request to it with the image and it sends it to slash predicted ABC, the local 66 00:04:42,190 --> 00:04:47,290 slash predict, you can send a file to our server. 67 00:04:48,010 --> 00:04:53,890 And once it receives a file, it executes this function here and it returns a class. 68 00:04:54,610 --> 00:04:56,130 So let's see this in action. 69 00:04:56,140 --> 00:04:57,730 So let's run this block of code. 70 00:04:57,730 --> 00:05:00,520 It takes maybe about five to 10 seconds to actually quicker. 71 00:05:01,060 --> 00:05:03,500 So it's running here on this link. 72 00:05:03,850 --> 00:05:10,400 That's so API link and you can see we just need to go to to visit the safe site because it doesn't have 73 00:05:10,400 --> 00:05:11,650 an SSL certificate. 74 00:05:12,380 --> 00:05:13,780 And this happens within. 75 00:05:13,780 --> 00:05:15,630 You have to refresh and then we go. 76 00:05:15,640 --> 00:05:20,470 So no starts running is just this is the it's not an actual webpage yet. 77 00:05:20,470 --> 00:05:21,220 It's an API. 78 00:05:21,700 --> 00:05:29,070 So now to accessto API, what we need to do, we need to open a new cloud notebook to open this one. 79 00:05:29,080 --> 00:05:29,890 That's our client. 80 00:05:34,430 --> 00:05:37,670 And let's download test image here of a bird. 81 00:05:38,840 --> 00:05:47,690 And let's get our previous and Gerlach link here pasted into their. 82 00:05:49,930 --> 00:05:54,700 Keeping this stash predict and let's see if we get to class, and there we go. 83 00:05:54,940 --> 00:05:56,770 So we've got the image here. 84 00:05:57,490 --> 00:06:01,010 So this image that you've downloaded want to visualize the image quickly? 85 00:06:01,010 --> 00:06:02,020 We can see it here. 86 00:06:02,620 --> 00:06:03,640 This cute little bird. 87 00:06:04,570 --> 00:06:09,250 So we send this picture out to the EPA and returns here. 88 00:06:09,400 --> 00:06:16,720 And so, no, you can actually post this from any machine across the world using this link, using any 89 00:06:16,810 --> 00:06:22,840 requests or any any sort of like the EPA tool interfacing tool we can do to any language you want. 90 00:06:23,380 --> 00:06:29,980 Or you can use something like postman to test these APIs and you get to know we just don't need this 91 00:06:29,980 --> 00:06:32,620 line here, but we get the class in return. 92 00:06:32,950 --> 00:06:38,590 So we get a class I.D. and the image, the class name lorikeets, a bid, by the way, in case you did 93 00:06:38,590 --> 00:06:41,440 not know, I wasn't sure it was until I googled it. 94 00:06:41,480 --> 00:06:44,020 We can actually google it to prove it. 95 00:06:45,160 --> 00:06:47,200 And you can see it's a type of parrot. 96 00:06:48,490 --> 00:06:49,180 Very pretty one. 97 00:06:49,570 --> 00:06:53,260 It's not exactly the same kind of bird, but it's very close as you can see. 98 00:06:53,290 --> 00:06:55,210 So we know our model is working well. 99 00:06:56,380 --> 00:06:58,060 So that's it for this lesson. 100 00:06:58,630 --> 00:07:06,790 In the next lesson, we're going to use Flask and Grok to create an actual website that allows you to 101 00:07:06,790 --> 00:07:12,520 access it, and you can accept some anywhere in the world and allows you to upload a picture and returns 102 00:07:12,520 --> 00:07:13,660 to class back to you. 103 00:07:14,200 --> 00:07:18,800 So let's stop there and we'll begin that flask web app section. 104 00:07:18,850 --> 00:07:20,020 No, thank you.