1
00:00:00,300 --> 00:00:00,870
Oh Ronnie.

2
00:00:01,110 --> 00:00:10,080
So now we've got our training split and validation split our X values are the file parts to our images

3
00:00:10,650 --> 00:00:11,700
and our y values.

4
00:00:11,700 --> 00:00:16,420
The labels are boolean num pi representations of the labels.

5
00:00:16,740 --> 00:00:23,710
And so what we have to do now is since our labels are already in a numeric format we have to convert

6
00:00:23,710 --> 00:00:28,500
how images into numbers or at least create a function to do so for us.

7
00:00:29,010 --> 00:00:41,030
So let's create a little heading for that will make it pre processing images turning images into tenses

8
00:00:42,630 --> 00:00:48,600
so we're kind of officially getting started on this step here and now when I say turning images into

9
00:00:48,600 --> 00:00:51,080
tenses This is a common theme.

10
00:00:51,630 --> 00:00:54,600
You'll have no matter what kind of data you're working on.

11
00:00:54,600 --> 00:01:01,170
So even if it's not images if it's something like an audio file or it's a text passage you're still

12
00:01:01,170 --> 00:01:06,840
going to have to turn it into tenses if you're working with tense flow and remember tenses are a numerical

13
00:01:06,930 --> 00:01:10,750
representation kind of like a Matrix All right.

14
00:01:10,950 --> 00:01:16,460
So what's one way we could do this again rather than pre processing our images in the notebook.

15
00:01:16,470 --> 00:01:25,560
We might pre process them only when we need to because manipulating file past is a lot easier than manipulating

16
00:01:25,560 --> 00:01:28,190
an image because it's working with less data.

17
00:01:28,280 --> 00:01:32,640
So because we've already got the file parts we're going to keep this theme for the rest of this project

18
00:01:32,760 --> 00:01:39,560
and I'm saying might do we're gonna do this we're gonna start developing out some functionality for

19
00:01:39,570 --> 00:01:44,910
a steps in the project and what I mean by that is creating functions rather than just writing one line

20
00:01:44,910 --> 00:01:47,250
of code and then running the cell.

21
00:01:47,370 --> 00:01:49,520
The cell won't run because there's nothing there.

22
00:01:49,710 --> 00:01:54,210
In the beginning when I first started out that's what I would do I would just run countless lines of

23
00:01:54,210 --> 00:01:59,940
code in countless cells and then keep running it and then eventually even if I was way down below in

24
00:02:00,390 --> 00:02:03,570
another part of the notebook I would have to rewrite the same code.

25
00:02:03,690 --> 00:02:09,090
I'd already written somewhere up here so that's why we're gonna start creating functions so that we

26
00:02:09,090 --> 00:02:15,210
can write code once it might take a little bit of time to begin with but we can reuse those functions

27
00:02:15,210 --> 00:02:19,110
later on and that's what I want you to stop thinking about in your workflows if you're thinking hey

28
00:02:19,800 --> 00:02:22,200
I'm going to be pretty processing the data here.

29
00:02:22,200 --> 00:02:27,240
Maybe I might need that function that functionality later so I'll turn it into a function rather than

30
00:02:27,240 --> 00:02:28,800
just a one off cell.

31
00:02:29,010 --> 00:02:37,300
So let's do that to pretty process our images into tenses.

32
00:02:37,300 --> 00:02:43,840
We're going to write a function which does a few things.

33
00:02:43,930 --> 00:02:50,020
Now this pre processing function will be specific to images but the concept remains the same.

34
00:02:50,230 --> 00:02:55,030
Whatever data you're working with you're probably going to have to turn it into some form of tensor.

35
00:02:55,030 --> 00:02:59,820
So we want to take an image file path as input.

36
00:02:59,830 --> 00:03:02,760
That's what we've got so far we've got image file parts.

37
00:03:02,890 --> 00:03:15,200
We want to use tensor flow to read the file and save it to a variable We'll probably call that variable

38
00:03:15,300 --> 00:03:15,940
image.

39
00:03:16,010 --> 00:03:17,530
Nice and simple.

40
00:03:17,870 --> 00:03:27,620
3 We want to turn our image so the image variable which will be a JPEG because that's the type of image

41
00:03:27,740 --> 00:03:34,930
image file that we're dealing with in two tenses for which we do next we need to raise size the image.

42
00:03:35,000 --> 00:03:41,600
So because all of our images are different sizes it's standard practice to usually resize all of your

43
00:03:41,600 --> 00:03:42,110
images.

44
00:03:42,110 --> 00:03:48,340
Image classification or some kind of deep pipeline with images into a standard eye size.

45
00:03:48,530 --> 00:03:55,820
So we'll resize the image to be a shape of 2 24.

46
00:03:55,820 --> 00:04:00,530
Now I want you to have to think about it without even having hands on experience with reshaping an image

47
00:04:00,530 --> 00:04:05,360
why might it be important to have it now this size can be arbitrary.

48
00:04:05,360 --> 00:04:10,610
But why might it be important to have images of a similar size when you're working with them.

49
00:04:10,670 --> 00:04:21,090
Finally return the modified image now we've got five steps there for the function you might be thinking

50
00:04:21,090 --> 00:04:22,600
wow how can I do that.

51
00:04:22,650 --> 00:04:24,830
Well I want you to have a look at resources.

52
00:04:24,880 --> 00:04:28,800
So you tend to flow data loading

53
00:04:31,990 --> 00:04:32,610
it we go.

54
00:04:33,110 --> 00:04:40,920
So I'll leave this as an aside that you can start to read of how to load data into intensive flow.

55
00:04:40,980 --> 00:04:45,180
So this is a really good section to check out so tense of flow guide data.

56
00:04:45,210 --> 00:04:48,660
I'll leave that in the resources section as an attachment if you want to check it out.

57
00:04:48,660 --> 00:04:51,200
I just spend 10 minutes reading through here.

58
00:04:51,300 --> 00:04:55,620
Maybe you could spend 10 minutes over the next couple of days even if you don't understand it.

59
00:04:55,620 --> 00:05:02,720
Just starting to get some of the concepts together and now as for tutorials there should be another

60
00:05:02,720 --> 00:05:03,250
one here.

61
00:05:04,450 --> 00:05:06,370
I think load and pre process data.

62
00:05:07,330 --> 00:05:14,580
So this is another valuable one we come here so this is what we're gonna be doing loading images.

63
00:05:14,580 --> 00:05:18,650
My problem is trying to load images and I want to know how to turn them into tenses.

64
00:05:18,780 --> 00:05:26,400
So I go to tutorials and look up load data and I'd look up how to load images and then you could read

65
00:05:26,400 --> 00:05:27,030
through here.

66
00:05:29,390 --> 00:05:32,150
So I'll make sure that I link these two resources.

67
00:05:32,150 --> 00:05:37,970
And after you finish this video I want you to read through them for about 10 minutes each so let's go

68
00:05:37,970 --> 00:05:41,780
back and let's build this function out here.

69
00:05:41,780 --> 00:05:47,930
But before we do I just want to give you a demonstration of what an image actually looks like if we

70
00:05:47,930 --> 00:05:51,200
were to read it in and what a tensor looks like.

71
00:05:51,200 --> 00:05:55,230
So let's convert an image to an umpire right.

72
00:05:58,500 --> 00:06:00,480
Right here before we do

73
00:06:04,050 --> 00:06:12,190
let's say what importing an image looks like so convert to an umpire right.

74
00:06:12,200 --> 00:06:19,540
Remember how I said tenses are very similar to multi-dimensional no high rise or matrices.

75
00:06:19,710 --> 00:06:20,560
So what we're gonna say now.

76
00:06:20,590 --> 00:06:25,870
So we're importing just to map what label function m raid which stands for image raid.

77
00:06:25,870 --> 00:06:34,060
We're gonna save image to m raid let's get file names 42 remember file names is just a list of file

78
00:06:34,060 --> 00:06:47,510
parts to our training images which is up here dog vision training images and then we want to check image

79
00:06:47,520 --> 00:06:48,000
dot shape

80
00:06:51,420 --> 00:06:53,760
OK so this is our image here.

81
00:06:53,760 --> 00:06:55,650
This is height width.

82
00:06:55,650 --> 00:07:01,770
Actually I always get confused but that's why we kind of reshape them to be like a square image so that

83
00:07:01,770 --> 00:07:03,450
height and weight doesn't really matter.

84
00:07:03,450 --> 00:07:10,320
I believe it's height with color channel and what I mean by color channel is red green and blue pixels.

85
00:07:10,380 --> 00:07:13,560
So if we go here what is a color.

86
00:07:13,560 --> 00:07:13,980
Channel

87
00:07:18,280 --> 00:07:25,400
Channel big color sample so this is what an image is made up of red green and blue colors.

88
00:07:25,400 --> 00:07:29,330
I'll let you dive a little bit more into that yourself but you're just going to have to trust me that

89
00:07:29,810 --> 00:07:35,660
each pixel and an image here is some combination of red green and blue values.

90
00:07:35,660 --> 00:07:39,680
And they range from 0 to 255.

91
00:07:39,890 --> 00:07:47,840
So if we have a look at our image which is a number higher Ray it should be.

92
00:07:47,920 --> 00:07:48,730
There we go.

93
00:07:48,760 --> 00:08:01,860
Values between 0 and 255 so if we go image dot Max maybe this will tell us 255 and image dot men

94
00:08:05,030 --> 00:08:05,880
0.

95
00:08:05,880 --> 00:08:06,540
There we go.

96
00:08:06,540 --> 00:08:09,620
So red green and blue pixel values between 0 and 2 5 5.

97
00:08:09,630 --> 00:08:15,000
That is literally what our image is now.

98
00:08:15,020 --> 00:08:16,310
It's in an umpire right.

99
00:08:16,310 --> 00:08:21,160
At the moment I remember how I said tenses are very similar to num high rise.

100
00:08:21,230 --> 00:08:24,180
So let's turn this into a tense and we can do that.

101
00:08:24,260 --> 00:08:28,880
You can turn almost anything into a tensor by using T F constant.

102
00:08:28,880 --> 00:08:36,910
Now if we read the doctoring creates a constant tensor from a tense a like object beautiful you can

103
00:08:36,910 --> 00:08:38,070
read a little bit more into that.

104
00:08:38,080 --> 00:08:40,750
But remember our motto if in doubt run the code.

105
00:08:40,750 --> 00:08:41,980
So that's we're going to practice.

106
00:08:42,520 --> 00:08:45,250
Let's see what this does.

107
00:08:45,280 --> 00:08:57,490
We've got the exact same values actually let's demonstrate this we want image and maybe only up to two.

108
00:08:57,550 --> 00:08:59,380
Same with TAF constant.

109
00:09:00,130 --> 00:09:08,050
So the only difference here is that now we've got it in a tensor of shape to three fifty three and now

110
00:09:08,080 --> 00:09:09,850
its shape to 353.

111
00:09:09,880 --> 00:09:12,070
We've indexed it to only show two.

112
00:09:12,160 --> 00:09:18,200
But if you see this is an array number why this is a tensor.

113
00:09:18,430 --> 00:09:25,000
The only difference here is that this tensor is intensive low and now it's able to run on a GP.

114
00:09:25,030 --> 00:09:26,280
So a lot faster compute.

115
00:09:26,290 --> 00:09:32,920
So GPA will be out to find patterns in these numbers a lot faster than what LCP you which is on my current

116
00:09:32,920 --> 00:09:40,820
machine my MacBook Pro will be out to find patterns in these let's go here turn image into a tensor.

117
00:09:41,270 --> 00:09:47,300
So now we've seen a little bit of the basics behind what an image is composed of what kind of data pixel

118
00:09:47,300 --> 00:09:50,300
values red green on blue between 0 2 5 5.

119
00:09:50,300 --> 00:09:56,010
We've seen what an image looks like in a non pi array and we've seen what an image looks like in a tensor.

120
00:09:56,060 --> 00:10:02,180
Let's finally get to our function which has to do these five steps.

121
00:10:02,180 --> 00:10:05,340
So what we might do to save this video from getting too long.

122
00:10:05,540 --> 00:10:09,890
I know I said out we were going to check out a write a function but I think it's good to build some

123
00:10:09,890 --> 00:10:11,680
intuition before we actually do right.

124
00:10:11,690 --> 00:10:17,230
This function will build this function in the next video so in the meantime try and put one of your

125
00:10:17,230 --> 00:10:18,830
own images here.

126
00:10:19,000 --> 00:10:20,650
You can choose any file name number.

127
00:10:20,710 --> 00:10:24,940
And then just have a play around with the different values and try turning it into a tensor.