WEBVTT - generated by wenglor-media

1
00:00:02.560 --> 00:00:04.400
To get started, we first need to connect

2
00:00:04.400 --> 00:00:06.720
weHub with an AI Lab account.

3
00:00:09.400 --> 00:00:10.200
To do this,

4
00:00:10.200 --> 00:00:12.800
open the weHub settings and enter your

5
00:00:12.800 --> 00:00:14.800
AI Lab credentials.

6
00:00:22.480 --> 00:00:23.520
Once that’s done,

7
00:00:23.520 --> 00:00:27.360
the AI Lab status of the device changes from
“Not Connected”

8
00:00:27.360 --> 00:00:29.360
to “Not Registered”.

9
00:00:34.360 --> 00:00:35.560
By connecting the device

10
00:00:35.560 --> 00:00:38.360
to our AI Lab account under the actions tab,

11
00:00:38.400 --> 00:00:41.280
the AI Lab status updates to “Registered”.

12
00:00:41.280 --> 00:00:42.000
With that,

13
00:00:42.000 --> 00:00:45.240
the connection of the AI Loop is successfully
established

14
00:00:45.240 --> 00:00:47.280
and we can continue with the next step:

15
00:00:47.280 --> 00:00:49.720
creating a dataset in the AI Lab.

16
00:00:51.200 --> 00:00:51.880
To do this,

17
00:00:51.880 --> 00:00:54.000
go to the AI Lab landing page

18
00:00:54.000 --> 00:00:56.040
and click the “CREATE DATASET” button.

19
00:00:56.160 --> 00:00:58.480
First, we need to give the dataset a name,

20
00:00:58.480 --> 00:01:00.000
so we enter one.

21
00:01:00.000 --> 00:01:01.000
Next, we can add classes –

22
00:01:01.000 --> 00:01:04.000
as many as we like, but at least two are required.

23
00:01:11.000 --> 00:01:14.000
Once that’s done, we can go into the dataset

24
00:01:14.000 --> 00:01:16.000
and copy the dataset ID.

25
00:01:19.000 --> 00:01:20.000
In the next step,

26
00:01:20.000 --> 00:01:23.000
we set up the Module Image AI in

27
00:01:23.000 --> 00:01:24.000
wenglor uniVision 3.

28
00:01:24.000 --> 00:01:26.000
First, open the camera settings

29
00:01:26.000 --> 00:01:29.000
and make sure the trigger mode is set to “On”

30
00:01:29.000 --> 00:01:31.000
and the trigger source is set to “Software”.

31
00:01:32.000 --> 00:01:34.000
Then, add the Module Image AI

32
00:01:34.000 --> 00:01:36.000
from the toolbox to the project tree.

33
00:01:37.000 --> 00:01:38.000
Link the input image and enter

34
00:01:38.000 --> 00:01:40.000
the dataset ID from the dataset

35
00:01:40.000 --> 00:01:42.000
we just created in the AI Lab.

36
00:01:43.000 --> 00:01:46.000
When selecting the “Upload to weHub” checkbox,

37
00:01:46.000 --> 00:01:48.000
we can define further settings,

38
00:01:48.000 --> 00:01:51.000
for example, how many images we want to add

39
00:01:51.000 --> 00:01:54.000
and whether these images should have tags.

40
00:01:54.000 --> 00:01:56.000
Tags help us distinguish good images

41
00:01:56.000 --> 00:01:58.000
from bad ones directly at the machine.

42
00:01:59.000 --> 00:02:01.000
So first, we upload 50 good images

43
00:02:01.000 --> 00:02:03.000
and assign them the tag “OK”.

44
00:02:03.000 --> 00:02:05.000
Once everything is configured,

45
00:02:05.000 --> 00:02:08.000
return to the camera settings and, under “Acquisition
Control”,

46
00:02:09.000 --> 00:02:11.000
switch the trigger mode back to “Off”

47
00:02:11.000 --> 00:02:14.000
so that the next 50 images will be automatically
uploaded

48
00:02:14.000 --> 00:02:18.000
to the AI Lab at the acquisition frame rate
that we’ve set.

49
00:02:20.000 --> 00:02:22.000
When we return to the AI Lab,

50
00:02:22.000 --> 00:02:27.000
we can see that our images are being uploaded
automatically.

51
00:02:30.000 --> 00:02:32.000
After uploading our first 50 good images,

52
00:02:32.000 --> 00:02:36.000
the Module Image AI in uniVision 3 informs
us

53
00:02:36.000 --> 00:02:38.000
that we need to reset the counter

54
00:02:38.000 --> 00:02:40.000
before uploading more.

55
00:02:41.000 --> 00:02:42.000
Before doing that,

56
00:02:42.000 --> 00:02:45.000
we adjust our setup to capture “NOK” images

57
00:02:45.000 --> 00:02:48.000
and enter this as a tag in the tag field.

58
00:02:49.000 --> 00:02:51.000
Once that’s prepared, we can reset the counter.

59
00:02:53.000 --> 00:02:55.000
From then on, the next 50 images are captured

60
00:02:55.000 --> 00:02:58.000
automatically and uploaded to weHub.

61
00:02:59.000 --> 00:03:01.000
After all images from both classes –

62
00:03:01.000 --> 00:03:04.000
okay and not okay – have been uploaded,

63
00:03:04.000 --> 00:03:06.000
we want to label them in batches.

64
00:03:06.000 --> 00:03:09.000
To do this, we use the filter options to select
the tags

65
00:03:09.000 --> 00:03:12.000
we previously defined in uniVision 3.

66
00:03:12.000 --> 00:03:15.000
First, we filter for our “ok” tags.

67
00:03:15.000 --> 00:03:17.000
Once the images are filtered,

68
00:03:17.000 --> 00:03:21.000
we select all of them and assign the ground
truth label “ok”.

69
00:03:24.000 --> 00:03:26.000
Then, we repeat the same

70
00:03:26.000 --> 00:03:28.000
process for the “nok” images,

71
00:03:28.000 --> 00:03:29.000
assigning them the corresponding

72
00:03:29.000 --> 00:03:32.000
ground truth label.

73
00:03:39.000 --> 00:03:40.000
Once all images in our dataset

74
00:03:40.000 --> 00:03:43.000
are labeled, we can start training our model.

75
00:03:43.000 --> 00:03:46.000
To do this, click the “TRAIN MODEL” button.

76
00:03:46.000 --> 00:03:48.000
On the model configuration page,

77
00:03:48.000 --> 00:03:50.000
we can choose the target device for deployment,

78
00:03:50.000 --> 00:03:53.000
either B60 or MVC.

79
00:03:53.000 --> 00:03:56.000
We can also define the image size for training
and select

80
00:03:56.000 --> 00:03:58.000
whether the images should be rectangular

81
00:03:58.000 --> 00:04:00.000
or square.

82
00:04:07.000 --> 00:04:09.000
When satisfied with the configuration,

83
00:04:09.000 --> 00:04:10.000
click “TRAIN MODEL”

84
00:04:11.000 --> 00:04:13.000
and the training process begins.

85
00:04:15.000 --> 00:04:17.000
When the model training is finished,

86
00:04:17.000 --> 00:04:20.000
the status on the model page will indicate
completion.

87
00:04:20.000 --> 00:04:22.000
Next, return to the

88
00:04:22.000 --> 00:04:25.000
Module Image AI in uniVision 3

89
00:04:25.000 --> 00:04:27.000
and open the AI model dialog.

90
00:04:27.000 --> 00:04:30.000
Here, select “Load Primary Model”.

91
00:04:30.000 --> 00:04:33.000
The model we just created is automatically

92
00:04:33.000 --> 00:04:35.000
downloaded from the AI Lab into uniVision 3.

93
00:04:36.000 --> 00:04:39.000
In the dialog window, a short preview of the
model

94
00:04:39.000 --> 00:04:41.000
and some additional information is displayed.

95
00:04:44.000 --> 00:04:45.000
In the software,

96
00:04:45.000 --> 00:04:46.000
we immediately

97
00:04:46.000 --> 00:04:48.000
see how the model’s predictions change depending

98
00:04:48.000 --> 00:04:50.000
on which image we select.

99
00:04:52.000 --> 00:04:53.000
Thank you for watching.