WEBVTT

00:00.120 --> 00:01.120
Welcome back.

00:01.240 --> 00:04.880
We validate and test our model.

00:05.160 --> 00:08.080
And this is the result on a new image.

00:08.120 --> 00:15.280
Now let's export the model as Tflite in order to use it in Android Studio.

00:15.320 --> 00:15.800
Okay.

00:15.960 --> 00:18.480
So create a new cell here.

00:18.600 --> 00:23.160
In order to export it as Tflite we need to use the TensorFlow.

00:23.160 --> 00:27.240
So install ultra and TensorFlow.

00:27.280 --> 00:30.000
TensorFlow from ultra.

00:31.120 --> 00:32.840
Import yolo.

00:32.960 --> 00:33.800
Create.

00:33.840 --> 00:45.320
This model equals to YOLO and you can copy the previous model copy selection run okay so this is from

00:45.320 --> 00:51.040
runs detect train runs detect train waits.

00:51.040 --> 00:54.720
And best this is our model okay.

00:54.760 --> 00:58.920
We need to convert it to Tflite in order to convert it.

00:58.960 --> 01:01.920
Go and use model dot export.

01:02.120 --> 01:06.990
And here set the format Equals do the f light.

01:06.990 --> 01:14.870
We talked about this extension which is for Android and microcontrollers in light version.

01:14.910 --> 01:15.390
Okay.

01:15.550 --> 01:19.830
Let me run this cell and wait for the result.

01:19.870 --> 01:21.870
And congratulations guys.

01:22.110 --> 01:27.230
We succeeded in exporting this model into Tflite.

01:27.270 --> 01:35.270
Okay so let me check where is the the model here runs detect a train.

01:35.270 --> 01:36.030
Wait.

01:36.070 --> 01:36.830
Saved.

01:36.870 --> 01:42.310
You see there is a new file and a new folder.

01:42.430 --> 01:43.910
This is best dot Onnx.

01:43.910 --> 01:45.710
We don't need the Onnx extension.

01:45.710 --> 01:46.990
We need the Tflite.

01:46.990 --> 01:49.510
So go and open best saved model.

01:49.510 --> 01:56.630
And here we have best Float16 Tflite and best Float32 Tflite versions.

01:56.670 --> 01:59.030
Okay, let me download both.

01:59.030 --> 02:01.350
And congratulations guys.

02:01.550 --> 02:07.750
We created the TF Lite version of our model.