1 00:00:00,05 --> 00:00:03,00 - [Instructor] Remember we used Teachable Machine, 2 00:00:03,00 --> 00:00:06,09 and we did a audio classification project. 3 00:00:06,09 --> 00:00:11,01 Now I'm going to show you image classification project. 4 00:00:11,01 --> 00:00:15,00 An image classification is a computer vision model. 5 00:00:15,00 --> 00:00:17,08 Remember the other computer vision model we learned? 6 00:00:17,08 --> 00:00:19,08 It's object detection, 7 00:00:19,08 --> 00:00:23,07 and we used Azure AI to do object detection. 8 00:00:23,07 --> 00:00:26,06 Now we are going to use Teachable Machine 9 00:00:26,06 --> 00:00:29,01 for image classification. 10 00:00:29,01 --> 00:00:32,08 So image classification and object detection 11 00:00:32,08 --> 00:00:35,02 are both computer vision models. 12 00:00:35,02 --> 00:00:37,03 What is the difference? 13 00:00:37,03 --> 00:00:38,07 If you remember, 14 00:00:38,07 --> 00:00:41,06 we started with object detection 15 00:00:41,06 --> 00:00:44,03 as our first demo in this course, 16 00:00:44,03 --> 00:00:50,04 and we fed the AI a simple image, 17 00:00:50,04 --> 00:00:55,00 and it put bounding boxes and identified objects. 18 00:00:55,00 --> 00:00:57,02 That's what object detection does; 19 00:00:57,02 --> 00:00:59,03 it takes a single image, 20 00:00:59,03 --> 00:01:01,01 we did not give a labeled image, 21 00:01:01,01 --> 00:01:04,02 we just gave an image, it is unlabeled, 22 00:01:04,02 --> 00:01:08,01 so in object detection, the image is unlabeled, 23 00:01:08,01 --> 00:01:12,09 and it identifies, the AI identifies, 24 00:01:12,09 --> 00:01:17,00 all the different objects and marks different locations 25 00:01:17,00 --> 00:01:19,08 where the objects are found, 26 00:01:19,08 --> 00:01:23,07 and it is trained using unsupervised learning, 27 00:01:23,07 --> 00:01:26,03 something like YOLO. 28 00:01:26,03 --> 00:01:31,03 Let's get started with image classification right here. 29 00:01:31,03 --> 00:01:32,09 With image classification, 30 00:01:32,09 --> 00:01:35,09 and I'm just going to select a Standard image model, 31 00:01:35,09 --> 00:01:40,01 in image classification, it is supervised learning. 32 00:01:40,01 --> 00:01:44,07 Supervised learning means we have to give labeled data. 33 00:01:44,07 --> 00:01:46,05 So we collect datasets 34 00:01:46,05 --> 00:01:48,09 that are going to be used as training data, 35 00:01:48,09 --> 00:01:54,00 but they need to be labeled, each image has a label. 36 00:01:54,00 --> 00:01:57,00 Unlike object detection, which is unsupervised, 37 00:01:57,00 --> 00:01:58,07 this is supervised learning. 38 00:01:58,07 --> 00:02:02,09 But with Teachable Machine, you don't need to worry. 39 00:02:02,09 --> 00:02:05,04 We are just going to find a dataset, 40 00:02:05,04 --> 00:02:07,05 and we are going to use it, 41 00:02:07,05 --> 00:02:09,09 and we are going to run machine learning 42 00:02:09,09 --> 00:02:12,03 with few simple clicks. 43 00:02:12,03 --> 00:02:15,01 Let me show you the datasets I'm planning to use, 44 00:02:15,01 --> 00:02:20,02 and I always use free datasets that are available 45 00:02:20,02 --> 00:02:24,08 and for easy download with no signups and no catch. 46 00:02:24,08 --> 00:02:27,04 So this Penn-Fudan dataset 47 00:02:27,04 --> 00:02:29,07 for Pedestrian Detection and Segmentation 48 00:02:29,07 --> 00:02:36,02 is a labeled dataset that we can use lot of images, 49 00:02:36,02 --> 00:02:39,02 and this is all people images. 50 00:02:39,02 --> 00:02:42,07 The other one is caltech-101 dataset, 51 00:02:42,07 --> 00:02:45,06 and this has close to 10,000 images, 52 00:02:45,06 --> 00:02:50,06 I think 9,000 plus images, in 101 object categories. 53 00:02:50,06 --> 00:02:53,01 When we download, we get all of them, 54 00:02:53,01 --> 00:02:56,07 and simple objects like elephant and chair and everything. 55 00:02:56,07 --> 00:02:59,03 And I'm just going to pick one category, 56 00:02:59,03 --> 00:03:01,06 and I'll show you what it is in a minute. 57 00:03:01,06 --> 00:03:05,03 So let's go back to Teachable Machine, and here is the URL. 58 00:03:05,03 --> 00:03:07,00 Follow along the demo, 59 00:03:07,00 --> 00:03:09,09 don't worry about the difference between object detection 60 00:03:09,09 --> 00:03:11,09 or image classification. 61 00:03:11,09 --> 00:03:15,02 I'm going to give you a handout which has all the details, 62 00:03:15,02 --> 00:03:18,06 it has the URL for all the different datasets 63 00:03:18,06 --> 00:03:19,05 that you can use, 64 00:03:19,05 --> 00:03:22,02 and the steps that you can follow along. 65 00:03:22,02 --> 00:03:26,01 So in machine learning, we are doing classification. 66 00:03:26,01 --> 00:03:28,01 We have to take our data 67 00:03:28,01 --> 00:03:32,00 and group them or classify them into multiple classes. 68 00:03:32,00 --> 00:03:35,00 And so, I'm going to call this People, 69 00:03:35,00 --> 00:03:38,04 or I'll call it Person, okay? 70 00:03:38,04 --> 00:03:41,07 You can call it anything, it's just a name. 71 00:03:41,07 --> 00:03:44,04 This other class, you know what I'm going to do? 72 00:03:44,04 --> 00:03:47,04 Remember, we've been talking about autonomous vehicles 73 00:03:47,04 --> 00:03:50,09 and identifying people and everything on the road, 74 00:03:50,09 --> 00:03:52,08 so I'm going to stick with that theme, 75 00:03:52,08 --> 00:03:55,09 though our Edge AI class is about Edge AI, 76 00:03:55,09 --> 00:03:58,06 about variety of AI running on the Edge 77 00:03:58,06 --> 00:04:00,04 for variety of different devices, 78 00:04:00,04 --> 00:04:02,07 I'm going to stick with the car example. 79 00:04:02,07 --> 00:04:06,07 And, typically, when autonomous vehicles 80 00:04:06,07 --> 00:04:08,03 are driving around, 81 00:04:08,03 --> 00:04:11,04 they identify all kinds of objects in their environment, 82 00:04:11,04 --> 00:04:14,03 and it is trained using object detection. 83 00:04:14,03 --> 00:04:16,04 But, occasionally, 84 00:04:16,04 --> 00:04:21,01 you might want to spot a specific item or an object 85 00:04:21,01 --> 00:04:22,01 and identify it. 86 00:04:22,01 --> 00:04:24,01 You can imagine in a road, 87 00:04:24,01 --> 00:04:26,06 it is not just about seeing whether there's a person 88 00:04:26,06 --> 00:04:30,02 on the road, which is good, right, pedestrian detection, 89 00:04:30,02 --> 00:04:33,07 but there are options in cars these days 90 00:04:33,07 --> 00:04:36,09 for identifying traffic lights or Stop signs. 91 00:04:36,09 --> 00:04:40,02 So I'm going to show you how we upload the data. 92 00:04:40,02 --> 00:04:41,04 Remember what we did? 93 00:04:41,04 --> 00:04:44,01 You simply press Upload, 94 00:04:44,01 --> 00:04:48,07 and you choose the images. 95 00:04:48,07 --> 00:04:53,03 And I have the PennFudanPad dataset downloaded. 96 00:04:53,03 --> 00:04:55,04 I literally went to the site I showed you, 97 00:04:55,04 --> 00:05:00,03 I clicked Download, and it gave me this directory. 98 00:05:00,03 --> 00:05:01,06 And I'm literally keeping 99 00:05:01,06 --> 00:05:04,00 all the different sub-directories that they showed, 100 00:05:04,00 --> 00:05:06,04 so that you know exactly what it looks like. 101 00:05:06,04 --> 00:05:09,02 I have not, you know, simplified or organized anything, 102 00:05:09,02 --> 00:05:11,05 so this is the same experience you will get 103 00:05:11,05 --> 00:05:13,03 when you download the dataset. 104 00:05:13,03 --> 00:05:17,00 And I need images, so I'm going to go to Images, 105 00:05:17,00 --> 00:05:20,07 and I'm going to get a bunch of images, 106 00:05:20,07 --> 00:05:25,00 let's say maybe 20 plus, let's see. 107 00:05:25,00 --> 00:05:26,06 22. 108 00:05:26,06 --> 00:05:32,00 So, you can see, these are all different people images. 109 00:05:32,00 --> 00:05:33,04 Now for Stop sign, 110 00:05:33,04 --> 00:05:36,01 do you remember what's the dataset we are going to use? 111 00:05:36,01 --> 00:05:39,05 We are going to use the caltech-101. 112 00:05:39,05 --> 00:05:42,01 So here I am at caltech-101, 113 00:05:42,01 --> 00:05:45,07 and as I told you, it has every possible object, 114 00:05:45,07 --> 00:05:47,00 not every possible object, 115 00:05:47,00 --> 00:05:49,09 but a lot of different object categories right here. 116 00:05:49,09 --> 00:05:53,01 And what do we need? We need the Stop sign. 117 00:05:53,01 --> 00:05:59,03 So I'm going to scroll down and find the Stop sign. 118 00:05:59,03 --> 00:06:01,01 It's right here. 119 00:06:01,01 --> 00:06:02,08 And if I click, 120 00:06:02,08 --> 00:06:05,04 now, it has a lot of different Stop signs. 121 00:06:05,04 --> 00:06:09,05 It's basically images of different Stop signs 122 00:06:09,05 --> 00:06:12,06 in different angles, in different lighting. 123 00:06:12,06 --> 00:06:14,08 So that's how different images are created 124 00:06:14,08 --> 00:06:16,02 for the same item. 125 00:06:16,02 --> 00:06:19,01 So I'm going to take a few from here, 126 00:06:19,01 --> 00:06:22,00 and I'm going to upload. 127 00:06:22,00 --> 00:06:25,08 Whoa, I got exact 22 images here too. 128 00:06:25,08 --> 00:06:27,03 That's amazing. 129 00:06:27,03 --> 00:06:29,01 Okay, so do you remember 130 00:06:29,01 --> 00:06:32,01 that we need to have almost similar size datasets 131 00:06:32,01 --> 00:06:35,02 for both the classes or multiple classes? 132 00:06:35,02 --> 00:06:37,04 It just happened, I got the exact 22. 133 00:06:37,04 --> 00:06:39,07 It could be 24 or 27, it doesn't matter, 134 00:06:39,07 --> 00:06:43,01 but it cannot be like 20 in one class 135 00:06:43,01 --> 00:06:45,02 and a hundred in another class, 136 00:06:45,02 --> 00:06:49,06 then you'll kind of skew the machine learning, 137 00:06:49,06 --> 00:06:51,08 or training of the model, 138 00:06:51,08 --> 00:06:55,00 to be biased towards the one with more data, okay? 139 00:06:55,00 --> 00:06:56,09 So now we have the Person dataset, 140 00:06:56,09 --> 00:06:59,01 we have the Stop sign dataset, 141 00:06:59,01 --> 00:07:00,09 what is the next step? 142 00:07:00,09 --> 00:07:05,06 I want to show you the epoch and the other factors in here. 143 00:07:05,06 --> 00:07:07,05 In audio, when we did the training, 144 00:07:07,05 --> 00:07:09,01 you remember the word epoch? 145 00:07:09,01 --> 00:07:13,09 This is the number of times the training AI 146 00:07:13,09 --> 00:07:16,04 will scan every single object, 147 00:07:16,04 --> 00:07:19,02 in this case, the images that we have given. 148 00:07:19,02 --> 00:07:21,00 And then there are other factors 149 00:07:21,00 --> 00:07:24,00 that you might want to just read if you're curious. 150 00:07:24,00 --> 00:07:27,03 But I'm just going to leave all the defaults here, 151 00:07:27,03 --> 00:07:30,09 and I'm just going to press Train. 152 00:07:30,09 --> 00:07:31,08 And done. 153 00:07:31,08 --> 00:07:35,05 You can see the 50 box that it's scanning for, okay. 154 00:07:35,05 --> 00:07:37,08 So the model is trained, 155 00:07:37,08 --> 00:07:39,02 voila, it's done. 156 00:07:39,02 --> 00:07:41,05 So now I'm going to choose an image. 157 00:07:41,05 --> 00:07:44,01 So here I have some samples we could use. 158 00:07:44,01 --> 00:07:45,07 Let's see. 159 00:07:45,07 --> 00:07:48,06 Remember we used these objects in our Azure demo? 160 00:07:48,06 --> 00:07:50,08 So maybe I want to try with that. 161 00:07:50,08 --> 00:07:53,04 Yes, these were all people and it says Person. 162 00:07:53,04 --> 00:07:56,03 It's perfect, it identifies people. 163 00:07:56,03 --> 00:08:00,01 I have an interesting one, a road scene, 164 00:08:00,01 --> 00:08:03,00 but I also want to test for the Stop sign first. 165 00:08:03,00 --> 00:08:06,00 So I'm going to go back to the caltech data, 166 00:08:06,00 --> 00:08:07,08 find the Stop sign. 167 00:08:07,08 --> 00:08:09,08 We used all these images, right, 168 00:08:09,08 --> 00:08:11,05 in the beginning, 22 of them. 169 00:08:11,05 --> 00:08:13,02 So I'm going to go down to an image 170 00:08:13,02 --> 00:08:16,00 that I have not used from the bottom, 171 00:08:16,00 --> 00:08:18,02 and it says Stop sign. 172 00:08:18,02 --> 00:08:19,06 The important thing is, 173 00:08:19,06 --> 00:08:23,05 we cannot use an image that is already being used 174 00:08:23,05 --> 00:08:25,04 in the training dataset. 175 00:08:25,04 --> 00:08:27,06 That's called overfitting. 176 00:08:27,06 --> 00:08:29,04 What do you think will happen? 177 00:08:29,04 --> 00:08:30,08 It's cheating. 178 00:08:30,08 --> 00:08:32,09 It's already got the answer, 179 00:08:32,09 --> 00:08:35,08 it has that in the brains of its learning, 180 00:08:35,08 --> 00:08:39,02 and we are just giving that as a test. 181 00:08:39,02 --> 00:08:41,08 So be careful to pick an image 182 00:08:41,08 --> 00:08:43,02 that has not been included 183 00:08:43,02 --> 00:08:46,05 in the training samples here, okay? 184 00:08:46,05 --> 00:08:51,00 So it identified a person, it identified a Stop sign. 185 00:08:51,00 --> 00:08:55,08 So I have an image from another dataset called Streetscenes, 186 00:08:55,08 --> 00:08:59,07 and let me input that and show you. 187 00:08:59,07 --> 00:09:01,05 See what happens here. 188 00:09:01,05 --> 00:09:06,00 So this is a lot of people and cars on a road. 189 00:09:06,00 --> 00:09:09,04 And, previously, when it identified people, 190 00:09:09,04 --> 00:09:10,07 it was very confident, 191 00:09:10,07 --> 00:09:13,08 when it identified a Stop sign, it was very confident. 192 00:09:13,08 --> 00:09:19,01 But now it is 95% confident that there are person in here, 193 00:09:19,01 --> 00:09:21,06 this label should be called a person, 194 00:09:21,06 --> 00:09:24,06 and it thinks maybe 5%, it's a Stop sign. 195 00:09:24,06 --> 00:09:26,01 Why do you think that's happening? 196 00:09:26,01 --> 00:09:28,06 There's no Stop sign in here. 197 00:09:28,06 --> 00:09:32,00 Because it has a lot of red, 198 00:09:32,00 --> 00:09:36,03 and the light by which it has learned Stop sign, 199 00:09:36,03 --> 00:09:39,06 it is a little confused thinking, 200 00:09:39,06 --> 00:09:44,02 "Maybe 5% confidence, I think there's a Stop sign in here," 201 00:09:44,02 --> 00:09:47,00 but it sees a lot of people it can recognize, 202 00:09:47,00 --> 00:09:49,08 so it thinks, "I'm going to place this as a label 203 00:09:49,08 --> 00:09:51,01 that is a person." 204 00:09:51,01 --> 00:09:54,04 So image classification 205 00:09:54,04 --> 00:09:58,00 is actually going to take an input image 206 00:09:58,00 --> 00:10:03,03 and it is going to say it belongs to this label as a person, 207 00:10:03,03 --> 00:10:06,03 or this label as a Stop sign. 208 00:10:06,03 --> 00:10:07,07 So that's all it is doing, 209 00:10:07,07 --> 00:10:12,03 it's literally labeling the image that we give. 210 00:10:12,03 --> 00:10:13,09 It's not like object detection 211 00:10:13,09 --> 00:10:17,00 to understand all the different elements in here. 212 00:10:17,00 --> 00:10:21,02 But every AI is not going to be 100% confident 213 00:10:21,02 --> 00:10:22,03 with every image. 214 00:10:22,03 --> 00:10:26,09 So it's important for you to run this model, train it, 215 00:10:26,09 --> 00:10:31,00 and test with a variety of different images. 216 00:10:31,00 --> 00:10:34,00 Think of the use case that you will use 217 00:10:34,00 --> 00:10:38,04 for image classification, you wanted to identify a thing. 218 00:10:38,04 --> 00:10:41,06 And in this case, the image classification happens 219 00:10:41,06 --> 00:10:44,01 because it takes an image that we have given, 220 00:10:44,01 --> 00:10:45,05 a bunch of images, 221 00:10:45,05 --> 00:10:48,00 and it understands, it's called feature extraction, 222 00:10:48,00 --> 00:10:49,09 it scans the image, 223 00:10:49,09 --> 00:10:54,04 it tries to understand the edges of the image, 224 00:10:54,04 --> 00:10:57,09 it tries to understand the texture of the image, 225 00:10:57,09 --> 00:10:59,04 it tries to understand 226 00:10:59,04 --> 00:11:02,03 what features contributes to this image, 227 00:11:02,03 --> 00:11:04,09 and that is why it knows a full image 228 00:11:04,09 --> 00:11:07,00 as that particular label. 229 00:11:07,00 --> 00:11:09,03 So when we have a complex image like that 230 00:11:09,03 --> 00:11:10,07 where there are people in there, 231 00:11:10,07 --> 00:11:13,01 it's maybe in the person class, 232 00:11:13,01 --> 00:11:15,00 but then it has this shiny red thing, 233 00:11:15,00 --> 00:11:17,05 it's never seen that before, 234 00:11:17,05 --> 00:11:20,08 so it's a little confused and makes a 5% mistake. 235 00:11:20,08 --> 00:11:24,06 So it's important for you to give lot of training data 236 00:11:24,06 --> 00:11:27,02 to give various different scenarios, 237 00:11:27,02 --> 00:11:29,04 various different lighting and shades 238 00:11:29,04 --> 00:11:32,09 and textures of the object you want to be identified. 239 00:11:32,09 --> 00:11:33,09 Think of this, 240 00:11:33,09 --> 00:11:38,06 if you want the car to identify a Stop sign and stop, 241 00:11:38,06 --> 00:11:41,06 then it needs to recognize it 242 00:11:41,06 --> 00:11:46,08 whether it is dark or whether it's in dim light, 243 00:11:46,08 --> 00:11:50,07 or maybe the Stop sign is in some snowy condition, 244 00:11:50,07 --> 00:11:52,08 and it's partially blocked, 245 00:11:52,08 --> 00:11:55,09 or it seems to have some kind of spots on it. 246 00:11:55,09 --> 00:11:59,00 So all kinds of situations 247 00:11:59,00 --> 00:12:03,01 which make the Stop sign look like it's something else 248 00:12:03,01 --> 00:12:04,07 should be fed as training data, 249 00:12:04,07 --> 00:12:06,09 and it's the same thing for a person 250 00:12:06,09 --> 00:12:09,00 or any other object you're trying. 251 00:12:09,00 --> 00:12:11,03 So with that, we are done with this demo. 252 00:12:11,03 --> 00:12:14,05 I hope you'll be able to try this with more other data 253 00:12:14,05 --> 00:12:16,09 from the different datasets that I shared 254 00:12:16,09 --> 00:12:18,06 to just compare it to different objects. 255 00:12:18,06 --> 00:12:22,02 But the important thing is not just this training, 256 00:12:22,02 --> 00:12:25,02 but it is about how you will use it. 257 00:12:25,02 --> 00:12:27,06 Oh, I need to show you how to export the model. 258 00:12:27,06 --> 00:12:30,06 So you have actually exported the previous model 259 00:12:30,06 --> 00:12:32,07 for Teaching Machine with voice, 260 00:12:32,07 --> 00:12:34,01 we are going to do exactly same thing. 261 00:12:34,01 --> 00:12:36,00 We clicked on Export Model, 262 00:12:36,00 --> 00:12:38,03 and one option is you can upload, 263 00:12:38,03 --> 00:12:41,00 and this will upload the model to the cloud, 264 00:12:41,00 --> 00:12:44,00 you copy it, and you have it, right? 265 00:12:44,00 --> 00:12:47,09 And you can also download the model and then you can use it, 266 00:12:47,09 --> 00:12:50,07 or you can get the JavaScript for this 267 00:12:50,07 --> 00:12:53,09 and you can give it to your engineer or data scientist, 268 00:12:53,09 --> 00:12:55,04 and you can incorporate this 269 00:12:55,04 --> 00:12:58,06 into many different applications in your workflow. 270 00:12:58,06 --> 00:13:00,02 So let me show you what that looks like. 271 00:13:00,02 --> 00:13:03,01 This is the URL where it went to the cloud. 272 00:13:03,01 --> 00:13:04,04 Think of where you would use this. 273 00:13:04,04 --> 00:13:06,07 If you want to show this to your customer 274 00:13:06,07 --> 00:13:09,02 about recognizes specific object, 275 00:13:09,02 --> 00:13:12,07 then this is the URL when you put that in the cloud, 276 00:13:12,07 --> 00:13:15,03 when you exported it and said Export 277 00:13:15,03 --> 00:13:16,09 and it saved it to the cloud, 278 00:13:16,09 --> 00:13:18,02 this is what it looks like. 279 00:13:18,02 --> 00:13:21,00 It'll not show you the whole training class and everything, 280 00:13:21,00 --> 00:13:24,03 it'll literally show you this window of the model, 281 00:13:24,03 --> 00:13:27,00 and you can click and input a file, 282 00:13:27,00 --> 00:13:29,05 and then show the output and show the model working, 283 00:13:29,05 --> 00:13:32,02 you can test this with your customer out in the field. 284 00:13:32,02 --> 00:13:34,07 But when you see this URL here, 285 00:13:34,07 --> 00:13:36,05 you also see links 286 00:13:36,05 --> 00:13:39,04 on how you can use these Teachable Machine models 287 00:13:39,04 --> 00:13:41,05 in your various projects. 288 00:13:41,05 --> 00:13:43,02 And that is very important, 289 00:13:43,02 --> 00:13:45,03 because if you're going to programmatically take 290 00:13:45,03 --> 00:13:47,05 all the content, how do you organize it? 291 00:13:47,05 --> 00:13:49,05 Where does it link? How do you do that? 292 00:13:49,05 --> 00:13:52,08 All this information is available in this link. 293 00:13:52,08 --> 00:13:54,06 So now we are truly done, 294 00:13:54,06 --> 00:13:58,01 and I'm calling on your imagination on what will you do? 295 00:13:58,01 --> 00:14:00,00 Don't think of this as two objects, 296 00:14:00,00 --> 00:14:03,01 think about how now you have learned two different models, 297 00:14:03,01 --> 00:14:04,05 you have learned object detection 298 00:14:04,05 --> 00:14:07,08 which identifies any kind of objects and person 299 00:14:07,08 --> 00:14:09,03 in a single image, 300 00:14:09,03 --> 00:14:12,00 and you've learned to do machine learning, 301 00:14:12,00 --> 00:14:13,04 image classification, 302 00:14:13,04 --> 00:14:17,01 so you can label items or you can take labeled datasets 303 00:14:17,01 --> 00:14:19,01 and now you can make a model, 304 00:14:19,01 --> 00:14:21,07 identify one thing confidently. 305 00:14:21,07 --> 00:14:25,00 So think about what kind of things would you like 306 00:14:25,00 --> 00:14:26,04 for this AI to build, 307 00:14:26,04 --> 00:14:29,05 so you can put it on any of your Edge devices, 308 00:14:29,05 --> 00:14:31,05 in any of your environment, 309 00:14:31,05 --> 00:14:34,00 and then make Edge AI work for you.