WEBVTT - generated by wenglor-media

1
00:00:03.280 --> 00:00:05.200
Now that our dataset is fully labeled,

2
00:00:05.200 --> 00:00:07.200
we are ready to train a model.

3
00:00:08.760 --> 00:00:10.160
Click “Train Model.”

4
00:00:10.160 --> 00:00:12.640
This opens the “Model Configuration” page.

5
00:00:12.800 --> 00:00:14.840
Here, you can give your model a name

6
00:00:14.840 --> 00:00:16.440
and choose the target device —

7
00:00:16.440 --> 00:00:19.040
either a B60 or an MVC.

8
00:00:19.320 --> 00:00:20.640
In the model architecture,

9
00:00:20.640 --> 00:00:23.080
choose “Latency” for faster performance

10
00:00:23.080 --> 00:00:25.280
or “Accuracy” for higher precision.

11
00:00:25.520 --> 00:00:27.720
You can also select the image “Image Size”.

12
00:00:28.240 --> 00:00:29.760
Only resolutions that can run

13
00:00:29.760 --> 00:00:31.960
on your selected hardware will be available.

14
00:00:32.400 --> 00:00:33.880
As a rule of thumb:

15
00:00:33.880 --> 00:00:37.120
Lower resolutions mean shorter training and
inference times,

16
00:00:37.600 --> 00:00:39.720
but if the resolution is too low,

17
00:00:39.720 --> 00:00:42.760
the model may struggle to distinguish between
classes.

18
00:00:43.480 --> 00:00:46.920
Try to choose a resolution that is as low as
possible,

19
00:00:46.920 --> 00:00:49.040
but as high as necessary.

20
00:00:49.520 --> 00:00:52.440
The “Resized Image Preview” gives you a
quick look

21
00:00:52.440 --> 00:00:55.400
at how your images will appear in that resolution.

22
00:00:56.040 --> 00:00:59.360
Below, you will also see the estimated training
time,

23
00:00:59.360 --> 00:01:02.000
inference time and the training cost in credits.

24
00:01:02.000 --> 00:01:04.000
If your classes are not rotation-dependent,

25
00:01:04.000 --> 00:01:06.000
you can enable image rotation (“Rotation”)

26
00:01:06.000 --> 00:01:08.000
for additional variation.

27
00:01:08.000 --> 00:01:10.000
Otherwise, it is best to keep it disabled.

28
00:01:10.000 --> 00:01:13.000
On the right side, you will find a summary
of your dataset —

29
00:01:13.000 --> 00:01:15.000
showing the number of classes,

30
00:01:15.000 --> 00:01:18.000
annotation coverage and dataset usage.

31
00:01:18.000 --> 00:01:19.000
Once everything looks good,

32
00:01:19.000 --> 00:01:22.000
click “Train Model” to start training.

33
00:01:22.000 --> 00:01:22.000
That’s it —

34
00:01:22.000 --> 00:01:24.000
your first model is now on its way!