Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00,090 --> 00:00:04,440
In the previous video, we have done training and have tried to detect face masks using the trained
2
00:00:04,440 --> 00:00:05,100
weights.
3
00:00:05,460 --> 00:00:10,860
In this video, we will measure the accuracy of the trained weights using mean average precision, mean
4
00:00:10,860 --> 00:00:15,180
average precision, or MFP is a metric for evaluating an object detection model.
5
00:00:15,480 --> 00:00:19,860
But before calculating MP, I will explain some of the arguments that can be used.
6
00:00:20,280 --> 00:00:21,860
First there is the weights argument.
7
00:00:21,870 --> 00:00:26,130
This is the goal of these seven weights file that will be used to calculate the MP value.
8
00:00:26,670 --> 00:00:28,830
This is an example of its application.
9
00:00:30,040 --> 00:00:32,140
Next is the bed size argument.
10
00:00:32,500 --> 00:00:35,650
This argument is the number of images processed at one time.
11
00:00:35,680 --> 00:00:37,960
This is an example of its application.
12
00:00:38,650 --> 00:00:40,390
Next is the device argument.
13
00:00:40,420 --> 00:00:44,950
This argument is used to select which CPU to use by writing down the index.
14
00:00:44,980 --> 00:00:49,920
The default value of this argument is zero, which means it selects the first available CPU with queue
15
00:00:49,960 --> 00:00:51,340
to support on the computer.
16
00:00:51,610 --> 00:00:57,250
If using CPU replace zero with CPU in this argument, the data argument comes next.
17
00:00:57,280 --> 00:01:02,200
This argument is a data file that contains the number of classes, the dataset product and the class
18
00:01:02,200 --> 00:01:02,680
name.
19
00:01:02,830 --> 00:01:05,080
This is an example of its application.
20
00:01:06,260 --> 00:01:07,860
Next is the IMT argument.
21
00:01:07,880 --> 00:01:10,580
This argument is the size of the image to be processed.
22
00:01:10,610 --> 00:01:13,070
The following is an example of its application.
23
00:01:14,460 --> 00:01:16,380
The confidence argument comes next.
24
00:01:16,740 --> 00:01:18,810
This is an object confidence threshold.
25
00:01:18,840 --> 00:01:21,360
The following is an example of its application.
26
00:01:21,450 --> 00:01:23,400
Next is the I.O.U. argument.
27
00:01:23,820 --> 00:01:25,400
This is the IOU threshold.
28
00:01:25,560 --> 00:01:30,390
I only use the ratio of the overlapping area between the predicted bounding box and the ground truth
29
00:01:30,390 --> 00:01:31,440
bounding box.
30
00:01:31,890 --> 00:01:36,760
The detection result is said to be correct if it has an IOU value greater than or equal to the threshold.
31
00:01:37,770 --> 00:01:39,930
This is an example of its application.
32
00:01:41,350 --> 00:01:42,760
Next is the name argument.
33
00:01:43,120 --> 00:01:47,320
This argument is the name of the folder that stores the P calculation results.
34
00:01:48,770 --> 00:01:50,960
This is an example of its application.
35
00:01:52,050 --> 00:01:53,910
Next is the Tusk argument.
36
00:01:54,450 --> 00:02:00,240
This argument specifies whether the calculations should be run on train validation or test data set.
37
00:02:01,430 --> 00:02:03,860
The following is an example of its application.
38
00:02:07,460 --> 00:02:12,170
In this video we will calculate upon validation and test of the face mask dataset.
39
00:02:16,480 --> 00:02:22,630
The validation dataset is located in the folder listed below and the test dataset is in the folder below.
40
00:02:27,130 --> 00:02:29,500
These are the images on the foundation dataset.
41
00:02:30,010 --> 00:02:31,420
There are 80 images.
42
00:02:34,910 --> 00:02:36,830
These are the images on the test dataset.
43
00:02:38,430 --> 00:02:39,930
There are 81 images.
44
00:02:46,490 --> 00:02:49,820
The first step to calculate MLP is to launch the Anaconda prompt.
45
00:02:50,570 --> 00:02:51,980
Press the Windows button.
46
00:02:53,200 --> 00:02:54,550
Then enter Anaconda.
47
00:02:54,580 --> 00:02:56,080
Click the Anaconda prompt.
48
00:03:00,530 --> 00:03:04,640
Then activate the all of seven CPU environment using the command.
49
00:03:05,620 --> 00:03:06,610
Activate yolo.
50
00:03:06,610 --> 00:03:08,560
v72 for you and for.
51
00:03:11,540 --> 00:03:12,350
Press enter.
52
00:03:16,480 --> 00:03:19,540
They never get to the goal of seven zips further.
53
00:03:29,590 --> 00:03:32,820
First, we will calculate MLP and the validation dataset.
54
00:03:33,080 --> 00:03:34,090
Use the command.
55
00:03:35,430 --> 00:03:37,350
Python test the PI.
56
00:03:37,440 --> 00:03:42,240
That's that's why it's we will use the train weights in this example in runs train.
57
00:03:43,130 --> 00:03:45,200
YOLO v seven Face mask.
58
00:03:46,130 --> 00:03:46,940
Waits.
59
00:03:47,950 --> 00:03:49,060
That's not pretty.
60
00:03:51,720 --> 00:03:53,550
In bedsides we write to.
61
00:03:56,930 --> 00:03:58,400
On device write zero.
62
00:04:02,890 --> 00:04:03,570
In the data.
63
00:04:03,610 --> 00:04:07,060
Write down the data file that was previously created in the training section.
64
00:04:10,760 --> 00:04:13,790
We use 640 pixels for the image size.
65
00:04:16,420 --> 00:04:21,399
In conference, we use 0.01, which is the default value for measuring accuracy.
66
00:04:24,470 --> 00:04:26,690
In IOU, we use 0.5.
67
00:04:30,740 --> 00:04:34,520
We write to all of his seven face must well in the name argument.
68
00:04:37,580 --> 00:04:39,080
In the last argument, right?
69
00:04:39,080 --> 00:04:42,230
Well, because we will calculate MLP in validation dataset.
70
00:04:45,050 --> 00:04:45,880
Chris enter.
71
00:04:51,790 --> 00:04:54,340
Wait until the calculation is finished.
72
00:05:02,770 --> 00:05:04,070
Here are the results.
73
00:05:04,090 --> 00:05:07,570
The calculation results will display precision recall.
74
00:05:08,510 --> 00:05:09,470
And LP.
75
00:05:10,180 --> 00:05:15,720
MP 0.5 indicates that the MLP calculates and employs an IOU threshold of 0.5.
76
00:05:16,180 --> 00:05:18,160
The detection result is said to be correct.
77
00:05:18,190 --> 00:05:21,320
If it has an IOU value of at least 0.5.
78
00:05:22,000 --> 00:05:26,470
In this example, the MLP value for all classes is 0.767.
79
00:05:28,610 --> 00:05:31,160
There are also MFP values for each class.
80
00:05:31,490 --> 00:05:35,990
This value can be used to determine whether the training results are suitable for all classes.
81
00:05:37,010 --> 00:05:42,320
In this example, the training results are good for mosque and no mosque, but not good for bad mosques.
82
00:05:45,300 --> 00:05:48,150
Next, we will calculate MLP on the test dataset.
83
00:05:48,390 --> 00:05:50,820
Use the common python.
84
00:05:50,820 --> 00:05:51,930
Test the python.
85
00:05:52,950 --> 00:05:54,240
That's that's why it's.
86
00:05:54,840 --> 00:05:56,520
We will use the train weights.
87
00:05:59,600 --> 00:06:01,460
In bed size we write through.
88
00:06:04,680 --> 00:06:06,120
On device write zero.
89
00:06:08,240 --> 00:06:08,920
In the data.
90
00:06:08,960 --> 00:06:12,410
Write down the data file that was previously created in the training section.
91
00:06:15,240 --> 00:06:18,270
We use 640 pixels for the image size.
92
00:06:20,110 --> 00:06:22,750
In contrast, we use 0.01.
93
00:06:26,450 --> 00:06:28,660
You know, you use 0.5.
94
00:06:32,660 --> 00:06:36,530
You write Your love is seven face must test in the name of human.
95
00:06:38,170 --> 00:06:42,640
In the first argument right test because we will calculate MLP in test dataset.
96
00:06:44,680 --> 00:06:45,490
Press enter.
97
00:06:49,630 --> 00:06:52,180
Wait until the MP calculation is finished.
98
00:07:00,400 --> 00:07:04,120
The following is the result of the map calculation on the test dataset.
99
00:07:07,250 --> 00:07:11,150
That's all explanation for measuring accuracy using mean average precision.
100
00:07:11,180 --> 00:07:11,780
Thank you.
101
00:07:11,780 --> 00:07:12,620
And see you then.
8487
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.