WEBVTT

00:00.080 --> 00:01.000
Welcome back.

00:01.040 --> 00:05.120
We added the required dependencies for our project.

00:05.280 --> 00:08.560
Now let's start building the data classes.

00:08.600 --> 00:09.920
Create a new package.

00:09.920 --> 00:11.360
Name it as data.

00:11.560 --> 00:16.000
Inside this package, create a new Kotlin data class.

00:16.120 --> 00:19.040
Name it as detection object.

00:19.040 --> 00:24.800
This data class has three variables val score.

00:24.840 --> 00:28.840
It's of type float val label.

00:28.840 --> 00:34.200
It's of type string val bounding box.

00:34.480 --> 00:40.160
It's of type rect f rect f from Android graphics.

00:40.160 --> 00:45.120
Inside this data package, create a new data class.

00:45.120 --> 00:48.520
Name it as detection parameters.

00:48.520 --> 00:56.040
Inside this detection parameters set val score threshold.

00:56.160 --> 01:00.440
It's of type float and set it to 0.6.

01:00.760 --> 01:11.480
So any object with inference with approximation greater than 60% go and display this object.

01:11.480 --> 01:13.080
So this is the threshold.

01:13.280 --> 01:17.400
Everything above 0.7 or 60%.

01:17.440 --> 01:19.320
Go and display this object.

01:19.320 --> 01:24.760
So we we trust that the object is 60% accurate.

01:25.120 --> 01:30.720
And this is a good a range and a good percentage for object detections.

01:30.720 --> 01:37.040
You can increase the inference if you need objects with higher inference and scores.

01:37.320 --> 01:47.320
So you set it to 0.8 which means 80% and detections and probability 0.9 is higher as and so on.

01:47.640 --> 01:48.080
Okay.

01:48.320 --> 01:51.200
I prefer using 0.66.

01:51.360 --> 01:52.960
And this is good.

01:52.960 --> 01:55.360
This is a score threshold.