WEBVTT 00:00.080 --> 00:00.960 Welcome back. 00:01.160 --> 00:02.200 Congratulations. 00:02.200 --> 00:04.160 We created our data set. 00:04.440 --> 00:06.160 You can train the model. 00:06.200 --> 00:07.160 Customer training. 00:07.160 --> 00:08.080 Quick training. 00:08.120 --> 00:11.040 Upload custom weights roboflow instant model. 00:11.080 --> 00:13.640 But we are not going to use that. 00:13.680 --> 00:17.440 The models that are are being created from Roboflow. 00:17.600 --> 00:23.560 We need to create our custom and train our model using Python. 00:23.560 --> 00:30.840 So I'm going to show you how to custom, how to create your custom object and how to create your custom 00:30.840 --> 00:39.080 training, and how to download the data set from Roboflow and import it into our Google Colab and use 00:39.080 --> 00:41.480 this data set to train our model. 00:41.640 --> 00:46.520 And we're going to train it with YOLO models. 00:46.840 --> 00:52.880 Okay, so go to virgins and download data set. 00:52.960 --> 00:55.560 We had to select a format. 00:55.800 --> 01:06.310 You can select a JSON, Coco, XML, Pascal, VOC if you are training with XML, txt yolo version 11, 01:06.350 --> 01:14.230 12, nine, and so on CSV, others and ultra Linux Hub and other Microsoft Azure. 01:14.390 --> 01:23.030 Okay, we in this section are interested in using YOLO version eight. 01:23.070 --> 01:23.750 Nine. 01:23.790 --> 01:24.510 11. 01:24.550 --> 01:25.070 12. 01:25.070 --> 01:25.950 And so on. 01:25.950 --> 01:29.950 So we're going to train our model using YOLO. 01:29.990 --> 01:30.470 Okay. 01:30.630 --> 01:32.870 So what is YOLO. 01:32.910 --> 01:35.950 We're going to discover in the next couple of minutes. 01:35.950 --> 01:39.190 But here select YOLO version eight. 01:39.350 --> 01:42.790 Don't use download zip to computer use. 01:42.830 --> 01:51.030 Show download code because we're going to use Google Colab to import and download our our data set into 01:51.070 --> 01:52.390 our Google Drive. 01:52.430 --> 01:52.870 Okay. 01:52.990 --> 01:54.670 So this is the code. 01:54.710 --> 01:56.670 This is our data set. 01:56.670 --> 02:05.030 We can copy this code and paste it in Google Colab to import and download our data set. 02:05.070 --> 02:07.150 Let's go to Google Colab. 02:07.310 --> 02:09.430 Create a new notebook. 02:09.610 --> 02:18.130 And by the way, you can download or download those notebooks from the resources folder under this section. 02:18.170 --> 02:26.090 Okay, so I'm gonna name it as YOLO Custom Object Detection okay. 02:26.130 --> 02:28.770 Click enter and click save. 02:28.890 --> 02:38.370 And as I told you you can find them in Python training and I'll name it as nine nine YOLO Custom Object 02:38.410 --> 02:40.370 Training in order to distinguish it. 02:40.370 --> 02:40.850 Okay. 02:41.090 --> 02:45.970 So the file name is called YOLO Custom Object detection okay. 02:46.330 --> 02:52.170 Now the first step is to download the data set from Roboflow. 02:52.170 --> 03:03.770 In order to download it we need to install pip install Roboflow and from roboflow import Roboflow. 03:03.970 --> 03:08.010 And here we need to paste this code okay. 03:08.170 --> 03:09.490 So copy this code. 03:09.650 --> 03:15.440 And by the way we have the code includes the pip install roboflow. 03:15.440 --> 03:17.640 So there's no need to write this okay. 03:17.800 --> 03:23.200 And you see that dataset is downloaded as version eight. 03:23.240 --> 03:28.440 Before running this code, let's learn about YOLO. 03:28.480 --> 03:32.520 YOLO stands for You Only Look Once. 03:32.520 --> 03:41.520 It's a revolutionary object detection system that that treats detection as a single unified regression 03:41.520 --> 03:42.440 problem. 03:42.680 --> 03:51.840 Unlike older systems that look an image an image multiple times in different regions, YOLO looks at 03:51.840 --> 04:00.160 the entire image just once to predict what objects are present and where they are. 04:00.360 --> 04:02.040 It's very fast. 04:02.200 --> 04:04.840 Think of it like like traditional methods. 04:04.840 --> 04:12.920 For example, sliding window like scanning a page with a magnifying glass, checking every small spot 04:12.960 --> 04:16.680 one by one to see if it contains an object. 04:16.880 --> 04:19.900 This is slow and inefficient. 04:20.060 --> 04:30.260 But when using YOLO, YOLO like glancing at the entire page of a split second of a split for a split 04:30.260 --> 04:36.900 second and immediately telling you all the words and their locations. 04:37.060 --> 04:39.460 This is incredibly fast. 04:39.500 --> 04:42.460 Okay, so YOLO is very fast. 04:42.500 --> 04:49.740 The input image is divided into a grid, for example 19 by 19 or s times s predict. 04:49.780 --> 04:58.540 Each grid cell is responsible for predicting bounding boxes, confidence score and class probabilities. 04:58.540 --> 05:05.820 Don't worry, we're going to learn about those in the next in the next couple of videos. 05:05.820 --> 05:09.820 But here I want from you to understand that YOLO is very fast. 05:09.860 --> 05:18.420 Looking at the image once and discover all the objects that are possible to be detected. 05:18.420 --> 05:22.700 So YOLO is extremely fast because it's a single stage detector. 05:22.850 --> 05:28.730 It can process videos in real time, 30 frames per second, enabling live object detection. 05:28.730 --> 05:34.370 We're going to see a quick example about this, a high accuracy global context. 05:34.370 --> 05:41.130 By looking at the whole image at once, it makes prediction with a better understanding of all of the 05:41.170 --> 05:47.210 other overall scene, leading to a fewer false positives in the background. 05:47.530 --> 05:51.850 Okay, YOLO is not a single model, but a family of models. 05:51.850 --> 05:57.010 It has been continuously improved by the research community. 05:57.210 --> 05:59.530 Versions like YOLO version. 05:59.570 --> 06:08.570 Version two, three, four and so on have pushed the boundaries of speed and accuracy. 06:08.730 --> 06:17.930 In short, YOLO look once approach made real time object detection practices, and it's very simple 06:18.250 --> 06:25.930 and very fast for object detection, powering countless applications from self-driving cars to medical 06:26.090 --> 06:26.850 imaging. 06:27.110 --> 06:32.350 okay to use the the ultra lytics. 06:32.350 --> 06:37.150 And and by the way, the ultra analytics is the company that invented the YOLO. 06:37.350 --> 06:43.550 And here you can search on Google YOLO, go to Doc's dot com. 06:43.550 --> 06:48.590 Or you can simply go to GitHub and search for ultra Lytics YOLO. 06:48.590 --> 06:54.870 Okay, so this is the GitHub GitHub Ultralytic Ultralytic slash actions. 06:54.870 --> 06:58.710 And you see that the YOLO okay. 06:58.870 --> 07:01.190 So this is the ultra Lytics. 07:01.190 --> 07:04.990 You can import it using pip install ultra Lytics. 07:05.150 --> 07:11.990 And in order to use YOLO you can use from ultra import YOLO and load the pre-trained model. 07:11.990 --> 07:15.190 This is a quick introduction about YOLO. 07:15.190 --> 07:24.190 In the next video, we're gonna learn about how to use YOLO in order to train our data set from Roboflow 07:24.190 --> 07:26.110 and start creating the model. 07:26.110 --> 07:29.230 The custom model that will detect the batteries.