WEBVTT 00:00.080 --> 00:08.680 We downloaded the data set into the folder Object Detection Android, and we imported it into our notebook. 00:08.920 --> 00:12.400 Now we're going to train the model. 00:12.440 --> 00:18.960 At first we need to edit this file called data dot YAML or y. 00:20.320 --> 00:26.360 Okay here we have the names only one class which is the battery. 00:26.720 --> 00:28.280 It's from Roboflow. 00:28.440 --> 00:35.320 And here we need to edit the test train and val variables. 00:35.360 --> 00:43.440 Let me get the test images copy path and here change the test to this. 00:43.640 --> 00:45.520 Also go to the train. 00:45.520 --> 00:51.920 Copy the Path of images folder from the train folder and paste it here. 00:52.040 --> 00:58.840 Also for the valid copy the images, copy path and paste it like this. 00:58.960 --> 01:02.250 Click save and close the file. 01:02.250 --> 01:06.210 So we added it this data dot YAML for a file. 01:06.250 --> 01:09.570 Now we're going to train our model. 01:09.850 --> 01:10.330 Okay. 01:10.530 --> 01:20.650 So our um our uh content folder contains only one file folder sorry which is object detection. 01:20.650 --> 01:21.370 Android. 01:21.410 --> 01:21.890 Okay. 01:22.210 --> 01:23.690 Uh, forget the sample data. 01:23.730 --> 01:27.210 This is a ready made one for every project. 01:27.490 --> 01:31.490 Uh, we created this object detection android. 01:31.530 --> 01:34.730 Okay, now let's train our model. 01:34.810 --> 01:43.130 At first, we need to install the ultra Lytics in order to use one of those, uh, one of those, um, 01:43.410 --> 01:44.930 YOLO versions. 01:44.930 --> 01:47.530 And we're going to use, uh, s or N. 01:47.690 --> 01:49.650 This is best for Android. 01:49.810 --> 01:55.810 So pip install ultra lytics click run and here we go. 01:56.130 --> 01:57.330 We execute it. 01:57.370 --> 01:58.970 Now scroll down. 01:59.010 --> 02:06.310 Create a new code cell and from ultra lytics import flow. 02:06.350 --> 02:10.070 Here, let me make this annotation. 02:10.190 --> 02:20.950 Load a model YOLO version eight not YAML, it's PT, so make sure you copy it if we go and open YOLO 02:20.950 --> 02:27.190 version YOLO 11 and it's YOLO 11 and PT okay. 02:27.350 --> 02:36.270 Also there are uh, the there are a lot of versions you can see in the page of ultra Lytics. 02:36.430 --> 02:44.310 So, uh, it's the best and the most used one is YOLO version eight. 02:44.350 --> 02:44.830 Okay. 02:45.150 --> 02:52.710 So there are a lot of versions we're gonna use YOLO version eight, which is a better and well-tested. 02:52.710 --> 02:57.550 So YOLO version eight and which is nano.pt. 02:57.910 --> 03:01.280 PT is the model extension. 03:01.280 --> 03:03.400 Now let's use this model. 03:03.640 --> 03:08.080 Results equals to model dot train. 03:08.120 --> 03:15.200 Now that train that train function needs a trainer and arguments. 03:15.440 --> 03:23.200 So here we need to pass the data equals to the content data YAML. 03:23.200 --> 03:24.440 So copy this. 03:24.680 --> 03:31.320 Copy the path of this file because it's very important from our object detection Android as. 03:31.320 --> 03:41.640 And as I told you we need to edit this file to correctly pass the images of validation images of training 03:41.640 --> 03:43.480 and images of testing. 03:43.480 --> 03:46.880 So make sure you edit this file before using it. 03:47.080 --> 03:50.320 Otherwise many problems will occur. 03:50.360 --> 04:00.780 Okay, so copy the path of it and paste it and pass it as the data of that data have the data parameter 04:00.780 --> 04:03.020 here of this train function. 04:03.060 --> 04:03.580 Okay. 04:03.780 --> 04:07.060 How many, uh, epochs we need? 04:07.100 --> 04:09.700 We need to, well, train our model. 04:09.700 --> 04:13.540 So I'm going to use the apex uh, 100. 04:13.820 --> 04:15.940 Also the image size. 04:15.940 --> 04:19.380 As I told you, we use 600 by 40. 04:19.420 --> 04:30.220 If we go to this table, the size will result will be, uh, will be in 640 times 640 height times width. 04:30.380 --> 04:30.860 Okay. 04:31.020 --> 04:38.580 And use the device I'm going to use, uh, if you have the GPU, you can select it one. 04:38.580 --> 04:41.460 Otherwise we're going to use the CPU. 04:41.700 --> 04:43.140 So device equals to zero. 04:43.140 --> 04:52.380 Also you can change the runtime uh to use a TPU, TPU and T4 GPU or you can use the CPU. 04:52.580 --> 04:57.950 It's not a big deal if you have a big data set, always use the GPU. 04:58.190 --> 04:59.470 Okay, run. 04:59.470 --> 05:04.590 This is, uh, this is resulting from, uh, changing the runtime. 05:04.590 --> 05:13.430 So if you change the runtime, just click run all to, to run all the previous codes and then execute 05:13.430 --> 05:13.950 this. 05:13.990 --> 05:14.470 Okay. 05:14.750 --> 05:21.990 So when changing the runtime, all the previous codes will be uh, will be removed or resetted. 05:22.030 --> 05:25.590 Um, and you should execute them again. 05:25.590 --> 05:26.030 Okay. 05:26.070 --> 05:30.070 So this is the first one now in training the data set. 05:30.070 --> 05:32.790 And here let's try to remove the device. 05:33.070 --> 05:37.310 If you have uh just the CPU run again. 05:37.310 --> 05:38.230 And here we go. 05:38.430 --> 05:45.630 So I removed the device because I'm not using the GPU of uh, the, the Google notebook. 05:45.870 --> 05:51.790 And please, guys, don't exit because we are executing here. 05:51.990 --> 05:54.150 So this is the first epic. 05:54.150 --> 05:55.950 It will be created here. 05:56.070 --> 06:00.000 And you see guys starting training for 100 epochs. 06:00.000 --> 06:07.480 Lugging results will be created to the content runs detect train and this is the first epoch. 06:07.760 --> 06:12.040 The image size is this and wait for the epochs. 06:12.200 --> 06:22.280 Also there is a very important note here that if you uh, if you notice that the training is slow because 06:22.280 --> 06:28.880 we are using the CPU, if you change it to GPU, it's faster. 06:29.080 --> 06:30.960 So let me change it to this. 06:30.960 --> 06:34.160 Click D for GPU, click save. 06:34.400 --> 06:37.760 Now we need to restart the runtime. 06:37.800 --> 06:38.280 Okay. 06:38.520 --> 06:41.720 Again let me click run all. 06:41.880 --> 06:46.240 And here it's very simple to use the GPU. 06:46.280 --> 06:50.920 Using the device equals to zero using the GPU. 06:51.160 --> 06:57.940 So I want from you to understand that using GPU is faster and in order to use the GPU. 06:58.260 --> 07:00.780 Use the change runtime. 07:01.060 --> 07:02.220 Default GPU. 07:02.220 --> 07:06.180 And to use this GPU you should use devices. 07:06.540 --> 07:09.940 Device equals to zero parameter when training the model. 07:09.940 --> 07:17.180 And if you notice guys that it's very, very and much faster than before, we are training the model 07:17.180 --> 07:19.780 in a fastest way using the GPU. 07:20.020 --> 07:22.180 And congratulations guys. 07:22.180 --> 07:27.220 100 epochs completed in 0.03 hours. 07:27.220 --> 07:31.020 And the validation is here the Ultralytic. 07:31.220 --> 07:36.740 And you see guys we have the results saved to content. 07:36.740 --> 07:38.940 The content is this folder right. 07:39.020 --> 07:39.540 Ones. 07:39.900 --> 07:45.380 This is the newly created folder runs Detect and Train. 07:45.620 --> 07:49.300 So here is our model. 07:49.300 --> 07:51.340 And the results is here.