1 00:00:00,480 --> 00:00:01,830 Welcome back. 2 00:00:01,860 --> 00:00:10,750 I hope you have installed get us an intensive float in your system now as a practice project. 3 00:00:10,790 --> 00:00:20,140 We are going to create an image classifier we will be classifying this kind of images into 10 different 4 00:00:20,140 --> 00:00:31,380 categories such as T-shirts trousers pullover dresses bags boots etc. For this we are going to use a 5 00:00:31,380 --> 00:00:37,690 very famous database that is known as fashion eminence database. 6 00:00:37,820 --> 00:00:45,890 Here we have our own seventy thousand great skill images of 10 different fashion categories objects 7 00:00:47,480 --> 00:00:58,560 our training site will be off 60000 images which we are going to use to screen our model we have and 8 00:00:58,660 --> 00:01:07,860 another set of 10000 images which we will be using as a basis to evaluate the performance of our model. 9 00:01:07,870 --> 00:01:16,930 These images are in the form of printed by 28 pixel is squared and each pixel is represented on the 10 00:01:16,930 --> 00:01:21,760 gray skin on a scale of 0 to 255. 11 00:01:21,840 --> 00:01:29,100 The great thing about this database is that it is available within gave us and we can directly imported 12 00:01:29,100 --> 00:01:30,180 from get us. 13 00:01:30,180 --> 00:01:37,650 We don't have to upload a separate fight to access this database here you can see the 10 different objects 14 00:01:37,710 --> 00:01:39,780 that are present in this database 15 00:01:43,330 --> 00:01:46,990 now to access this database from gave us. 16 00:01:46,990 --> 00:01:54,850 We are first creating a fashion underscored and this object where we are calling this database and after 17 00:01:54,850 --> 00:02:01,110 that we are loading this database and our X and vice data sites. 18 00:02:01,160 --> 00:02:08,000 We are calling our train dataset as X underscore xn underscore the full and via underscore train underscore 19 00:02:08,010 --> 00:02:14,530 full and our test dataset as X underscore best and find the scored best. 20 00:02:14,570 --> 00:02:24,860 So just load this database now we can use my library to view images in this database. 21 00:02:24,860 --> 00:02:32,290 So for example if I want to excel the first image I can just write X underscore green underscore full 22 00:02:32,650 --> 00:02:42,660 and we are accessing the first element that is the picture that is present at position 0. 23 00:02:42,950 --> 00:02:45,800 You can see that this is our first image. 24 00:02:45,800 --> 00:02:54,590 If you want to access the second image I'm just changing the location to 1 in this way you can access 25 00:02:54,590 --> 00:03:00,710 the different images that are present in this database. 26 00:03:01,020 --> 00:03:03,860 Now this is our X video. 27 00:03:04,290 --> 00:03:08,630 This out of the pixels that we are going to use to predict the object. 28 00:03:10,540 --> 00:03:18,310 Now to view the actual category of this object we have to call the wide screen dataset so you can just 29 00:03:18,310 --> 00:03:21,760 call the element that is present at the first position. 30 00:03:21,760 --> 00:03:29,330 So if I run this you can see that the output is zero. 31 00:03:29,530 --> 00:03:38,290 To view the category that is responding to the Zero label we can refer to this about table here zero 32 00:03:38,290 --> 00:03:40,120 sense for t shirt and tops. 33 00:03:40,150 --> 00:03:51,110 One is sense or closer to sense for the lower and so on so this image is of a t shirt and the output 34 00:03:51,290 --> 00:03:53,760 is also representing that. 35 00:03:53,770 --> 00:03:54,900 This is a t shirt 36 00:03:57,960 --> 00:03:58,870 V. 37 00:03:58,880 --> 00:04:10,110 Just check this for the first element you can see that this is a boot and the white label is 9. 38 00:04:11,300 --> 00:04:20,870 If you see the label line correspond to ankle boot so instead of referring people each time we can create 39 00:04:20,870 --> 00:04:31,300 a list of last name where we have lists certain all the categories in the order of their labels so that 40 00:04:31,390 --> 00:04:37,340 if I call the first element of this list it frenetically Give me the t shirt. 41 00:04:37,450 --> 00:04:41,670 If I call the second element it will give closeups as an output. 42 00:04:42,370 --> 00:04:48,480 So instead of calling the labels I can directly call the description of those labels. 43 00:04:48,520 --> 00:04:58,370 Using this last name list so just check the image of object that this president look Sean Penn 44 00:05:02,590 --> 00:05:05,120 to check the white label of this object. 45 00:05:05,240 --> 00:05:13,890 I can ethically call last name and then the label of the pollution and you can see the last name. 46 00:05:13,890 --> 00:05:14,650 Here it is. 47 00:05:14,660 --> 00:05:15,540 These are total 48 00:05:19,180 --> 00:05:19,880 few notice. 49 00:05:19,880 --> 00:05:27,710 Here we are using my block to block the data that is stored in the extreme dataset not to view the content 50 00:05:27,710 --> 00:05:28,990 of this data. 51 00:05:29,120 --> 00:05:29,840 You can 52 00:05:33,550 --> 00:05:37,520 can just write extra in full and then call the object. 53 00:05:38,680 --> 00:05:45,380 So earlier I have mentioned that this images are of 28 white 28 grayscale format. 54 00:05:46,180 --> 00:05:53,250 So here in the data you are seeing 28 in 28 pixel values. 55 00:05:53,380 --> 00:06:02,270 These are the pixels that are present at the first rule these are the pixels that are present. 56 00:06:02,290 --> 00:06:10,750 The second row and so on for that can be a draw here zero represent pure black and blue fifty five represents 57 00:06:10,770 --> 00:06:11,490 white. 58 00:06:11,710 --> 00:06:19,200 So the location of first pixel that is the first row and the first pixel you can see it's pure black. 59 00:06:19,210 --> 00:06:22,050 That's why we are getting zero. 60 00:06:22,090 --> 00:06:27,250 So our data is present in this form to view the data. 61 00:06:27,250 --> 00:06:28,170 You have to use. 62 00:06:28,210 --> 00:06:36,540 I am sure mantle and to get the vibe and lose the actual category of this data you have to call white 63 00:06:36,570 --> 00:06:44,440 grained data and you can also use last name list to directly get the descriptions set off label. 64 00:06:44,440 --> 00:06:49,310 So this is the data that we are going to use for each record. 65 00:06:49,360 --> 00:07:00,310 We have 28 in 28 values that is 784 values and using these values we are going to predict the description 66 00:07:00,400 --> 00:07:03,460 of this that solve for the data. 67 00:07:03,460 --> 00:07:07,510 You have the raw data can view this data and you have the last names. 68 00:07:07,510 --> 00:07:12,350 Now we have to create model on this data in the next lecture. 69 00:07:12,370 --> 00:07:19,600 We will normalize our dataset and for divided our trained dataset into validation entering site thinking.