WEBVTT 00:05.140 --> 00:06.730 Hi everyone and welcome back. 00:06.920 --> 00:13.910 So in this new series of videos we're going to be implementing a new API which uses machine learning 00:13.910 --> 00:15.500 to classify an image. 00:15.770 --> 00:20.390 So the idea behind what we're going to do is we're going to be given an image. 00:20.420 --> 00:24.750 So let's say this is an image and was a person inside. 00:24.920 --> 00:27.440 So excuse my horrible drawing. 00:27.800 --> 00:32.320 But then the image would have some things that it might have an animal that might have an airplane. 00:32.330 --> 00:35.690 You know it might not have anything that might have even the SOMEONE WHO famous. 00:35.690 --> 00:39.840 For example Angelina Jolie or someone or some other person. 00:39.860 --> 00:40.570 OK. 00:40.640 --> 00:45.580 And then we're going to give this model this image into our API. 00:45.620 --> 00:48.590 The user is going to give this image into our API. 00:48.830 --> 00:50.740 And then we're going to do some machine learning. 00:50.780 --> 00:51.630 OK. 00:52.130 --> 00:57.330 And then we're going to classify this image into one of five categories. 00:57.380 --> 01:01.490 So we're going to tell the user hey this is probably a person. 01:01.940 --> 01:05.450 And 80 percent sure about that. 01:05.600 --> 01:12.140 This might be a pineapple and I am one percent sure about that. 01:12.260 --> 01:15.870 This might be a famous person OK. 01:17.010 --> 01:18.620 And that I'm sure about that. 01:18.630 --> 01:20.260 For example 15 percent. 01:20.420 --> 01:22.020 And this might be an animal right. 01:22.030 --> 01:29.910 Somehow somehow the model classifies as animal 3 percent and then I notice this might be something else 01:29.970 --> 01:31.350 x 4. 01:31.350 --> 01:34.800 So we have 80 percent and then 4 percent plus five. 01:34.800 --> 01:35.700 That's 19. 01:35.700 --> 01:39.960 So 19 plus eight is ninety nine so one percent is x. 01:39.970 --> 01:40.680 Ok. 01:40.800 --> 01:44.130 So are going to be given this image over here. 01:44.130 --> 01:47.050 We're going to do some machine learning using Genser flow. 01:47.130 --> 01:53.880 So Tanzer flow which is a library for deep learning that was published by Google. 01:53.910 --> 01:58.920 So we're not going to be training our own model or are you going to use one of the models that Google 01:58.920 --> 01:59.690 already trained. 01:59.690 --> 02:03.810 It's called N.S. v3. 02:04.000 --> 02:04.580 OK. 02:04.780 --> 02:11.290 So inception B-3 is a very strong model and is able to classify an image into one of these classes for 02:11.510 --> 02:13.420 those too many classifications. 02:13.420 --> 02:20.620 I'm going to show you the model in a few seconds but it can basically classify animals plants humans 02:20.650 --> 02:26.710 famous people airplanes you know non-human things and so on. 02:26.710 --> 02:32.800 So it's a really powerful model that we can just use automatically since Google already trained it so 02:32.800 --> 02:38.230 we can just use it somehow like an API you can think of it as we're using this as an API and we're going 02:38.230 --> 02:41.260 to see how we can use it in the coming video. 02:41.290 --> 02:46.350 And I'm also going to explain tensor flow a little bit just the workflow of how it flow works. 02:46.360 --> 02:53.290 So we are more you know we understand how everything would work basically in in our API. 02:53.800 --> 02:59.380 Now in this video we're going to first of all introduce it we introduce what we're here to do so we're 02:59.380 --> 03:07.250 going to be building an image classification API or image classification API. 03:08.050 --> 03:14.560 And in fact if you Google just an image classification API or image recognition API you're going to 03:14.560 --> 03:17.610 find hundreds of Abiah ised that or any do this. 03:17.740 --> 03:21.540 And you know they might have even used section B-3 under the hood. 03:21.580 --> 03:22.410 So you never know. 03:22.420 --> 03:23.040 Right. 03:23.050 --> 03:29.590 So in fact there are many startups actually doing this service as an API so they offer an API and so 03:29.590 --> 03:31.120 an API would be here. 03:31.120 --> 03:33.400 So the user would access this API. 03:33.490 --> 03:38.300 And then you would get a response of this result right away so he doesn't have to do all this machine 03:38.360 --> 03:40.520 learning or heavy lifting. 03:40.600 --> 03:41.410 If you mean it. 03:41.540 --> 03:42.480 OK. 03:42.550 --> 03:49.630 So yeah this is this is actually very much like real world scenarios where you might be asked to develop 03:49.630 --> 03:51.390 an API that does something like this. 03:51.430 --> 03:52.520 OK. 03:52.660 --> 03:57.040 All right so the second thing we're going to do in this video is to do a chart protocol right where 03:57.130 --> 04:01.730 we want to do a protocol off all the resources that we have. 04:01.780 --> 04:02.470 OK. 04:02.620 --> 04:05.460 So we're going to do this very very simple great rehab. 04:05.470 --> 04:08.500 We're going to have three things we're going to have to register. 04:08.890 --> 04:09.480 OK. 04:09.610 --> 04:12.880 And as you might have guessed it's just registration. 04:12.880 --> 04:19.810 We're also going to have to classify OK to classify an image. 04:19.810 --> 04:20.480 What is it. 04:20.500 --> 04:22.180 You know what's in that image. 04:22.240 --> 04:25.160 And then finally I'm going to also add refill. 04:25.180 --> 04:25.850 OK. 04:26.170 --> 04:32.620 So we've already done refill in a previous video but we're also going to do it again to basically. 04:32.650 --> 04:39.520 So when the user registers he gets let's say 5 tokens for free so he can use our classification API 04:39.520 --> 04:40.730 for five times. 04:40.960 --> 04:47.850 But then once he's out of these five he can pave basis and then in return the administrator of the EPA 04:47.850 --> 04:51.360 I can refill his points for this user. 04:51.900 --> 04:56.070 OK so these are the resources I should have written is resources here. 04:56.080 --> 04:57.680 Now what about the your l. 04:57.730 --> 05:04.860 Well L is pretty pretty straightforward right for a register is going to be slash register for classifieds 05:04.880 --> 05:07.090 going to slash classify. 05:07.480 --> 05:10.190 And for a refill it's going to be last refill. 05:10.360 --> 05:11.210 OK. 05:11.850 --> 05:12.340 All right. 05:12.340 --> 05:13.750 So what about the method. 05:13.760 --> 05:16.600 Or is it a get post what is it. 05:16.780 --> 05:19.510 Well for registration was actually right. 05:19.510 --> 05:20.730 The parameter is first. 05:20.770 --> 05:22.750 OK so are there any parameters. 05:23.020 --> 05:33.040 So for arrester I need the username and the password and all that user right for classify I need the 05:33.040 --> 05:40.240 user name password to verify that he's actually the one he he's the same username right. 05:40.240 --> 05:46.020 Or otherwise I can use someone else's user name and someone else's tokens to classify my own images 05:46.020 --> 05:46.740 right. 05:46.740 --> 05:51.900 So we're going to use the get the username password and we're also going to get it you are. 05:52.020 --> 05:56.050 OK so this you are going to be your out to the image. 05:56.070 --> 06:02.190 Right so it's age DGP something something to puji for example OK. 06:02.250 --> 06:04.280 And then this is the image on the Internet. 06:04.290 --> 06:07.620 And we're going to download it and classify that image for the user. 06:07.620 --> 06:08.400 OK. 06:09.090 --> 06:11.130 For the refill it's straightforward. 06:11.130 --> 06:15.760 We have the user name the admin password right. 06:15.780 --> 06:18.840 Because we don't want anybody to be able to refill this. 06:18.870 --> 06:25.350 Any user with any number of tokens right or otherwise you would be at the point of refilling right. 06:25.350 --> 06:30.060 You just keep refilling your account for it with an incident number off of tokens. 06:30.360 --> 06:31.610 And then finally the amount. 06:31.630 --> 06:35.420 OK how many points or how many tokens do I want to add to this user. 06:35.640 --> 06:36.540 All right. 06:37.020 --> 06:38.340 OK cool. 06:38.340 --> 06:40.220 Now let's talk about the method. 06:40.240 --> 06:43.100 You can see that all of them require parameters. 06:43.090 --> 06:45.640 Right we have to send something to the server. 06:45.750 --> 06:48.920 And so in these cases we're going to use posts for all of them. 06:48.990 --> 06:49.490 Right. 06:49.620 --> 06:52.620 Because post is when we want to send something to the server. 06:52.710 --> 06:56.680 So post post and post. 06:56.760 --> 06:57.330 All right. 06:57.330 --> 06:59.950 So the final thing is the return codes. 07:00.030 --> 07:00.590 OK. 07:00.630 --> 07:04.020 So the return codes were statuses and so on. 07:04.440 --> 07:12.110 So all of them or as you might have guessed they have all hands 200 OK 200 OK and 200 OK. 07:12.120 --> 07:12.650 Right. 07:12.690 --> 07:16.120 That's if nothing no problems happen. 07:16.290 --> 07:20.070 Now in registration we might have another problem we might have 3 0 1. 07:20.130 --> 07:28.620 And we discussed this three and one here would mean that user name already exists already exists. 07:28.620 --> 07:32.180 So it's an invalid username here. 07:32.250 --> 07:33.160 We might also. 07:33.160 --> 07:39.830 So so we can shorten this three to one as invalid username. 07:39.870 --> 07:41.140 OK. 07:41.940 --> 07:45.230 Again to classify it we might have three in one. 07:45.270 --> 07:49.790 Someone might send that username is ABC and there's no ABC in our database right. 07:49.790 --> 07:51.510 The user doesn't exist. 07:51.510 --> 07:55.800 So three or one invalid username. 07:55.970 --> 07:57.250 OK. 07:57.750 --> 08:00.490 What about the password username password. 08:00.510 --> 08:01.940 Well they might not match right. 08:01.940 --> 08:06.930 I might my Bashford might be secure but someone sends ABC. 08:06.930 --> 08:08.600 Right so it's not my password. 08:08.640 --> 08:11.370 So 1:57 is invalid. 08:12.130 --> 08:19.280 Password and then the final thing is three or three what if he classified wait why did the username 08:19.290 --> 08:21.050 is correct the password is correct. 08:21.210 --> 08:24.720 But hey you wants to classify an image but you doesn't have enough tokens. 08:24.750 --> 08:27.610 He needs to refill though he needs to pay first. 08:27.690 --> 08:35.130 So in this case we send them back to 3 0 3 and then we tell him sorry out of tokens again out of tokens. 08:36.560 --> 08:37.550 All right. 08:37.700 --> 08:39.010 Now for the refill. 08:39.020 --> 08:41.260 We have something similar for the username. 08:41.270 --> 08:46.380 Well you might put a user name who doesn't exist in our database so it's an invalid user him. 08:46.490 --> 08:50.570 So there are three or one invalid username. 08:50.660 --> 08:55.500 OK so that means that this user is not doesn't exist in our database. 08:55.500 --> 08:57.690 Another one might be three or four. 08:57.740 --> 09:03.080 So if I send an admin password if I'm a hacker trying to hack into the ABC I refill my account with 09:03.080 --> 09:05.410 an infinite number of tokens. 09:05.430 --> 09:09.230 Then if this admin password is correct then I won't need to donate. 09:09.230 --> 09:11.150 Sorry this is a wrong password. 09:11.150 --> 09:16.240 So three or four invalid and password. 09:16.300 --> 09:22.220 OK administrator password and then that's it for now. 09:22.290 --> 09:27.750 You can also you know do an extra thing where the amount has to be positive. 09:27.980 --> 09:33.440 But then again the only person who is going to be using this slash refill is the administrator of the 09:33.440 --> 09:34.130 API. 09:34.220 --> 09:38.100 So most probably he's not going to implode like negative 5 for the end. 09:38.200 --> 09:41.920 And he is most probably going to end but five he doesn't want to break the system. 09:42.200 --> 09:48.080 So usually these checks are invalidations you put them just to make sure that nothing happens or no. 09:48.100 --> 09:51.760 Nobody tried to keep your system basically secure. 09:52.070 --> 09:54.740 So this is basically the API for for it. 09:54.740 --> 09:57.900 So registration we already did that before and the Riggio were about. 09:57.920 --> 10:03.590 We already did that before so we should go through them fairly quickly without too much explanation. 10:03.670 --> 10:08.210 But for the classification here well this is the interesting point because we're going to be getting 10:08.240 --> 10:09.130 your l. 10:09.220 --> 10:15.590 Sorry going to download this image and then we're going to fly the machine learning model that Google 10:15.590 --> 10:19.960 already trained on it and then we get the results and we return it to the user. 10:20.000 --> 10:20.670 OK. 10:20.840 --> 10:22.960 So hopefully this made sense. 10:22.970 --> 10:28.330 So we're going to stop here and in the next video we're going to take off and start implementation of 10:28.340 --> 10:34.410 the entire you know structure of the folders and our files doc and compose and so on. 10:34.820 --> 10:36.100 So yeah we're going to stop here. 10:36.110 --> 10:38.180 And until the next video happy coding.