WEBVTT

00:00.040 --> 00:02.520
Hello developers, and welcome back.

00:02.560 --> 00:06.160
In this video we're going to learn about Teachable Machine.

00:06.360 --> 00:10.360
Go to Teachable Machine dot with google.com website.

00:10.360 --> 00:18.280
And this is an amazing website for training a computer to recognize your own images, sounds and poses.

00:18.280 --> 00:27.160
So Teachable Machine is a free web based tool developed by Google that makes it incredibly easy to create

00:27.160 --> 00:31.400
machine learning models without writing any code.

00:31.440 --> 00:33.040
Yes, you heard it right.

00:33.040 --> 00:39.280
It's incredibly easy to create ML projects and models without writing any code.

00:39.480 --> 00:46.920
It's designed for educators, students, artists, and anyone curious about AI.

00:47.240 --> 00:55.000
You can create models for image projects, image classification and object detection, audio projects,

00:55.000 --> 01:02.760
sound classifications, pose projects, body pose estimation, and for our application.

01:02.800 --> 01:06.600
Now in this section we're going to use it for object detection.

01:06.660 --> 01:15.780
So we're going to use like these modals to indicate for example hard disk and earphones and classify

01:15.860 --> 01:16.260
them.

01:16.300 --> 01:16.820
Okay.

01:17.020 --> 01:21.820
So we're going to use this amazing website to train our models.

01:22.100 --> 01:24.180
It's a very simple click.

01:24.180 --> 01:25.220
Get started.

01:25.260 --> 01:29.260
Of course you need to sign in with your Google account.

01:29.260 --> 01:36.100
And here you can open an existing project from drive, or open an existing project from a file.

01:36.180 --> 01:40.660
Or you can select Image Project, Audio Project, or Post project.

01:40.860 --> 01:46.220
In this section we're going to use Image Image Project.

01:46.340 --> 01:56.300
And here you can use standard Image model best for most uses color images 2 to 4 times to to to to for

01:56.460 --> 01:58.540
export to TensorFlow TF Lite.

01:58.540 --> 02:04.020
And this is our target to get the A flight models as we did in the previous sections.

02:04.020 --> 02:07.620
And it's around five megabyte embedded image model.

02:07.620 --> 02:18.360
It's a simple and it's um like a smaller export to TF Lite for microcontrollers TF Lite and TF dot js.

02:18.400 --> 02:20.360
Okay, so it's very simple.

02:20.360 --> 02:22.040
I'm going to use the standard.

02:22.040 --> 02:25.000
And here we have two classes.

02:25.000 --> 02:27.760
You can add much class as you want.

02:28.000 --> 02:39.680
So uh here in this section we're gonna indicate and classify the images or detect objects uh, using

02:39.840 --> 02:43.920
uh two classes the hard disk and earphones.

02:43.960 --> 02:44.440
Okay.

02:44.640 --> 02:48.560
So let me name this class as hard disk.

02:48.760 --> 02:52.880
And the second one is earphones or AirPods.

02:52.880 --> 02:54.320
And this is good okay.

02:54.360 --> 02:57.440
So we have two classes hard disk and AirPods.

02:57.480 --> 03:02.240
For the hard disk class we have a webcam and upload.

03:02.240 --> 03:04.880
Also this is for the AirPods.

03:04.880 --> 03:14.000
So you can open webcam and uh, start like training and visualizing the hard disk if you if you have

03:14.000 --> 03:15.440
any hard disk.

03:15.560 --> 03:23.460
Also you can train the model and upload the images from Google Drive or from your computer.

03:23.460 --> 03:26.820
So we have two options webcam and upload.

03:26.820 --> 03:31.260
I want from you to understand a very important thing.

03:31.420 --> 03:34.620
If you choose webcam here I don't have webcam.

03:34.780 --> 03:42.740
You need to point your webcam at the hard disk and take images from different different angles and different,

03:42.780 --> 03:44.460
um, backgrounds.

03:44.700 --> 03:47.780
Uh, smaller images, bigger big images.

03:47.820 --> 03:54.780
Zoom into the hard disk and make zoom in, zoom out and vary your samples.

03:54.780 --> 04:02.020
Hold your hard disk in different orientations, under different lighting, partially off camera and

04:02.020 --> 04:03.980
alongside other projects.

04:04.300 --> 04:09.860
Then hold the uh hold to record button to capture many images.

04:09.860 --> 04:19.620
For example, 100 to to 200 per class is a good start, so we need to get 100 to 200 images for each

04:19.660 --> 04:20.420
class.

04:20.540 --> 04:26.590
Repeat the process with the airport class, Untangle them and untangle them.

04:26.710 --> 04:28.230
Place them neatly.

04:28.470 --> 04:34.270
Have just have just one earbuds in the in the frame and so on.

04:34.310 --> 04:34.750
Okay.

04:34.950 --> 04:42.990
So you you need to get different orientations, different lighting, partially off camera and alongside

04:42.990 --> 04:48.790
with other objects in order to distinguish, um, that the objects.

04:49.070 --> 04:55.630
So as I told you, you have two ways the webcam and upload you if you have pictures for your, uh,

04:55.630 --> 05:00.150
hard disk or any class, you can upload them directly.

05:00.310 --> 05:07.150
But for making things very fast and, uh, training the model, I'm gonna use the webcam.

05:07.350 --> 05:09.310
As I told you, I don't have webcam.

05:09.510 --> 05:13.110
I have I need to use my cell phone.

05:13.110 --> 05:15.710
So, uh, go to this website.

05:15.750 --> 05:24.510
Ah, I download the webcam for windows and download it, download the application from Play Store or

05:24.550 --> 05:25.150
the App Store.

05:25.150 --> 05:30.730
If you have a iPhone and run the application and you get like this.

05:30.770 --> 05:33.890
Okay, so this is our, um, webcam.

05:34.370 --> 05:41.210
My Android device is showing the hard disk and the earphones and the AirPods.

05:41.250 --> 05:45.490
Okay, so those are the two objects I'm gonna detect.

05:45.490 --> 05:51.050
So go to the webcam and here allow when while visiting the site.

05:51.050 --> 05:54.970
And you can notice that everything is working fine.

05:55.090 --> 06:05.130
Now in order to record and save images, you, um, you need to use hold to record.

06:05.170 --> 06:06.410
So like this.

06:06.770 --> 06:09.650
Okay, so hold to record.

06:09.690 --> 06:12.850
I'm gonna I'm showing images like this.

06:12.850 --> 06:14.970
You see, guys, how many images?

06:14.970 --> 06:16.490
130 images.

06:16.490 --> 06:19.850
And you can flip the hard disk like this.

06:19.890 --> 06:26.970
Hold to record, change the directions and the zoom in.

06:27.010 --> 06:28.130
Zoom out.

06:28.250 --> 06:29.530
And here we go.

06:29.570 --> 06:32.530
Now see you guys like this okay.

06:32.770 --> 06:35.370
And different orientations okay.

06:35.410 --> 06:38.550
Those are 527 images.

06:38.870 --> 06:40.150
Those are good.

06:40.270 --> 06:42.190
Okay, you see them guys?

06:42.190 --> 06:43.110
It's good.

06:43.230 --> 06:44.910
Save samples to drive.

06:44.910 --> 06:48.670
If you want to save them to the drive and click okay.

06:48.950 --> 06:54.910
So those are the images for the first hard disks uh, class.

06:54.910 --> 06:58.910
Now let's use the same for the airports.

06:58.950 --> 07:06.550
Click webcam and hold the camera hold to record and start recording.

07:06.590 --> 07:07.150
Okay.

07:07.470 --> 07:10.350
So those are the images.

07:10.710 --> 07:13.790
And this is the EarPods.

07:13.950 --> 07:19.630
And so okay make 360 and continue okay.

07:20.030 --> 07:23.350
So here I'm getting the images continue.

07:23.470 --> 07:26.590
And this is good okay.

07:26.630 --> 07:31.190
So zoom in zoom out get different angles lighting and so on.

07:31.310 --> 07:35.070
Oops it's 9914.

07:35.110 --> 07:36.070
It's good.

07:36.110 --> 07:38.110
Click x and here we go.