1
00:00:00,540 --> 00:00:06,600
And welcome back in this section, we'll take a look at building an obstacle to using the ultra-Orthodox

2
00:00:06,600 --> 00:00:11,280
YOLO Vision five framework to train a blood cell detector.

3
00:00:11,550 --> 00:00:15,230
So we're going to detect different types of blood cells in these medical images.

4
00:00:15,240 --> 00:00:17,820
These slides of these microscope slides.

5
00:00:18,360 --> 00:00:19,620
So let's take a look.

6
00:00:19,710 --> 00:00:23,880
So let's open Notebook 51 here, and let's get started.

7
00:00:23,880 --> 00:00:26,630
So you can see this is how it's labeled.

8
00:00:26,640 --> 00:00:32,880
So you can see these cells here with this bluish tint to it that I guess that's a white blood cell.

9
00:00:33,300 --> 00:00:36,990
And these smaller ring looking ones are two red blood cells.

10
00:00:37,410 --> 00:00:39,060
And these are platelets here.

11
00:00:39,060 --> 00:00:44,130
So you can see those are the three main classes in this medical imaging dataset.

12
00:00:44,640 --> 00:00:46,520
So again, this is a rebel floor notebook.

13
00:00:46,530 --> 00:00:53,250
However, this one is actually provided by cultural ethics that mean removal to onto the text implementation

14
00:00:53,250 --> 00:00:54,510
implemented it here.

15
00:00:55,260 --> 00:00:56,970
So let's take a look at it now.

16
00:00:56,970 --> 00:01:01,200
I will tell you ultimately did an amazing job with your vision.

17
00:01:01,200 --> 00:01:01,560
Five.

18
00:01:02,220 --> 00:01:05,400
So feel free to visit the GitHub repo.

19
00:01:05,580 --> 00:01:07,980
Actually, we can take a look at it here.

20
00:01:08,940 --> 00:01:15,540
In just visit this link right here and we can see the GitHub repo.

21
00:01:18,860 --> 00:01:25,610
So you can see your image in five here, you have a cloud implementation, also, Cargill is a slack

22
00:01:25,610 --> 00:01:30,680
forum where you can ask the developers questions as a Docker image as well if you needed one.

23
00:01:31,910 --> 00:01:33,740
It's quite easy to install and setup.

24
00:01:33,740 --> 00:01:36,680
It works on CPU systems or GPU systems.

25
00:01:36,680 --> 00:01:39,140
However, don't trade it in the CPU system.

26
00:01:39,140 --> 00:01:42,710
It's going to take forever and probably going to overheat your machine.

27
00:01:44,220 --> 00:01:46,100
But you can see it's quite well integrated.

28
00:01:46,100 --> 00:01:51,740
It has a number of integrations with things like weeds and biases and other metrics.

29
00:01:52,370 --> 00:01:53,240
So it's quite good.

30
00:01:53,250 --> 00:01:54,340
Encourage you to use this.

31
00:01:54,380 --> 00:01:56,080
You can see it's quite popular on GitHub.

32
00:01:56,090 --> 00:01:58,490
It has 21000 stores as well.

33
00:02:00,090 --> 00:02:03,070
So I've already run around these blocks of code here.

34
00:02:03,130 --> 00:02:06,570
This is just to clone the repo and still the requirements.

35
00:02:07,290 --> 00:02:08,550
And then we get the data set.

36
00:02:08,580 --> 00:02:09,870
Download it here.

37
00:02:11,030 --> 00:02:17,570
And we just do some file configurations as well and print the configuration file.

38
00:02:17,950 --> 00:02:24,110
The Yamoah, actually they don't call to see if she filed in this report or two of six.

39
00:02:24,980 --> 00:02:30,350
And then we can just write some things to the file here, some changes we need to make and balances

40
00:02:30,350 --> 00:02:34,160
from number of classes, actually, which comes from an earlier function.

41
00:02:35,140 --> 00:02:37,560
And then we can just start training.

42
00:02:38,110 --> 00:02:39,370
So let's begin.

43
00:02:40,240 --> 00:02:42,890
So you can see your model has begun training here.

44
00:02:42,910 --> 00:02:43,590
Let's take a look.

45
00:02:43,600 --> 00:02:46,240
This is what you should see when you start training.

46
00:02:47,020 --> 00:02:52,330
You can see all of these things here, and that's good architecture that is displayed.

47
00:02:52,840 --> 00:02:57,340
And then you can see the training results after each epoch and you can see it's training quite fast

48
00:02:57,340 --> 00:02:57,730
anyway.

49
00:03:00,490 --> 00:03:06,580
Let's pause there for now, and I'll show you the inference for to see on the Vision five model using

50
00:03:06,580 --> 00:03:09,100
the blood cell detection dataset.

51
00:03:09,280 --> 00:03:11,590
So let me just pause this recording here.

52
00:03:17,940 --> 00:03:21,900
So now you have the performance metrics you can analyze with this tense on board here.

53
00:03:21,930 --> 00:03:27,510
You can also get the images for the results so you can take a look at the precision to recall the map

54
00:03:27,530 --> 00:03:29,700
because those are quite important as well.

55
00:03:30,450 --> 00:03:36,000
And you can take a look at inference on a few of the images here, so you can see this is the ground

56
00:03:36,000 --> 00:03:37,050
to data, first of all.

57
00:03:37,500 --> 00:03:42,060
And no, we're going to print the creating results.

58
00:03:42,060 --> 00:03:45,570
Yes, and this is the actual results on some of our test data.

59
00:03:46,350 --> 00:03:47,910
And you can see this is quite good.

60
00:03:48,270 --> 00:03:56,040
Now you can run it in a few random images here and take a look at them individually here for these images.

61
00:03:56,040 --> 00:03:58,980
And you can see this is also quite good.

62
00:03:58,980 --> 00:04:02,190
It looks quite similar to the other one, but that's because it is similar.

63
00:04:02,730 --> 00:04:07,890
So you can see this is performing quite well with your emission five and then you can export these weights

64
00:04:07,890 --> 00:04:10,620
for future use if you want to save them.

65
00:04:12,180 --> 00:04:14,920
So overall, I think this is quite good.

66
00:04:14,950 --> 00:04:16,350
Hope you're happy with this lesson.

67
00:04:16,890 --> 00:04:21,760
In the next lesson, we'll take a look at a plant doctor dataset.

68
00:04:21,780 --> 00:04:22,500
What is that?

69
00:04:22,530 --> 00:04:25,950
Well, it's looking at different plant lives for disease.

70
00:04:26,160 --> 00:04:28,270
It's quite a few different classes in it.

71
00:04:28,650 --> 00:04:34,470
And again, we're going to use the ultra takes, you know, information five pi to its implementation.

72
00:04:34,740 --> 00:04:36,150
So stay tuned for that lesson.

73
00:04:36,330 --> 00:04:36,720
Thank you.