1 00:00:02,000 --> 00:00:02,000 Hi guys. 2 00:00:02,000 --> 00:00:09,000 So now I will train my Yolo v eight model on the pen and box dataset which I have created in the previous 3 00:00:09,000 --> 00:00:09,000 lecture. 4 00:00:09,000 --> 00:00:11,000 You can see over here. 5 00:00:11,000 --> 00:00:18,000 So I will just export this dataset into my Google CoLab notebook by just copying this and pasting this 6 00:00:18,000 --> 00:00:24,000 into my Google CoLab notebook, I will be able to export my dataset from Roboflow into my Google CoLab 7 00:00:24,000 --> 00:00:25,000 notebook. 8 00:00:25,000 --> 00:00:31,000 So first of all, before we export the dataset from Roboflow into my Google CoLab notebook, we first 9 00:00:31,000 --> 00:00:39,000 need to have we need to install YOLO v eight so YOLO V8 can be implemented in two ways by installing 10 00:00:39,000 --> 00:00:44,000 the package Ultralytics or by cloning the YOLO V8 GitHub repo by. 11 00:00:44,000 --> 00:00:50,000 So YOLO V8 is the first version of the YOLO to have its own official package, so we will demand YOLO 12 00:00:50,000 --> 00:00:55,000 V8 by using PIP install ultralytics by installing the package by cloning. 13 00:00:55,000 --> 00:01:01,000 The GitHub repo is a bit lengthy process, so I will follow the easy way which is PIP install ultralytics 14 00:01:01,000 --> 00:01:03,000 to install yolo v8 and implement it. 15 00:01:03,000 --> 00:01:09,000 So first of all, I will set my home directory as this path and setting my current directory. 16 00:01:09,000 --> 00:01:16,000 And now I will install Android ethics package to implement YOLO V8 so it will install all the required 17 00:01:16,000 --> 00:01:17,000 libraries as well. 18 00:01:17,000 --> 00:01:22,000 So now we will check whether Ultralytics is imported and it's working fine or not. 19 00:01:23,000 --> 00:01:23,000 Okay. 20 00:01:23,000 --> 00:01:30,000 So until it is imported and it's working fine, so now I'm creating a folder over here by the name dataset, 21 00:01:30,000 --> 00:01:31,000 so I've already created it. 22 00:01:31,000 --> 00:01:35,000 So it's appearing cannot create directory because it's already exists. 23 00:01:35,000 --> 00:01:36,000 You can see over here. 24 00:01:37,000 --> 00:01:40,000 Okay, so I'm just checking what is my present working directory. 25 00:01:40,000 --> 00:01:48,000 So now I am setting my present working directory as this dataset so I can import my dataset from roboflow 26 00:01:48,000 --> 00:01:51,000 into this colab notebook datasets folder. 27 00:01:51,000 --> 00:01:52,000 Okay. 28 00:01:53,000 --> 00:01:59,000 So I will just copy this all from here and just paste it over here. 29 00:01:59,000 --> 00:02:02,000 Just remove this and paste it and run this. 30 00:02:02,000 --> 00:02:09,000 So now all data set will be downloaded from roboflow into my Google CoLab notebook into this datasets 31 00:02:09,000 --> 00:02:10,000 folder. 32 00:02:10,000 --> 00:02:13,000 So it might take some few minutes or few seconds. 33 00:02:13,000 --> 00:02:17,000 So the data stood for the dataset to get downloaded. 34 00:02:17,000 --> 00:02:17,000 Okay. 35 00:02:17,000 --> 00:02:18,000 What's the issue? 36 00:02:18,000 --> 00:02:19,000 Let me see. 37 00:02:19,000 --> 00:02:20,000 Okay. 38 00:02:20,000 --> 00:02:23,000 I think then we can copy it completely. 39 00:02:23,000 --> 00:02:27,000 So just click on export over here, show download and see. 40 00:02:27,000 --> 00:02:29,000 Okay, so now it will be downloaded. 41 00:02:30,000 --> 00:02:31,000 There. 42 00:02:32,000 --> 00:02:36,000 But if you face this error as well, just remove this and just run this. 43 00:02:36,000 --> 00:02:42,000 So now we will be able to export our dataset from Roboflow into our Google CoLab notebook. 44 00:02:43,000 --> 00:02:47,000 So if you ever face this issue, just go here and click on. 45 00:02:49,000 --> 00:02:52,000 Export again and just copy this one as well. 46 00:02:52,000 --> 00:02:53,000 So this error will be fixed. 47 00:02:53,000 --> 00:02:58,000 Okay, so now our dataset is being downloaded and it's completely downloaded. 48 00:02:58,000 --> 00:03:03,000 You can see that 100% downloaded and it's available in the datasets folder. 49 00:03:03,000 --> 00:03:07,000 You can see pan book detection and train test validation. 50 00:03:07,000 --> 00:03:10,000 And this is the data dot yml file. 51 00:03:10,000 --> 00:03:15,000 You can see that we have two classes Pan book and this is the train test validation path. 52 00:03:15,000 --> 00:03:19,000 Okay, so let's see, what is our dataset location? 53 00:03:19,000 --> 00:03:22,000 So we need to make few changes over here. 54 00:03:22,000 --> 00:03:30,000 Let me tell you, just rename this folder and just remove one because it because if there is any numeric 55 00:03:30,000 --> 00:03:31,000 number, dash one. 56 00:03:31,000 --> 00:03:34,000 So our model is unable to assess the dataset path. 57 00:03:34,000 --> 00:03:42,000 Now just go to read data, dot HTML and just go to train and just copy this path. 58 00:03:43,000 --> 00:03:46,000 And just paste it over here. 59 00:03:46,000 --> 00:03:51,000 Similarly, will go over here as well and write test over here. 60 00:03:52,000 --> 00:03:58,000 Okay, so I'm just setting the path for that train test and validation and write validate over here. 61 00:03:59,000 --> 00:03:59,000 Okay. 62 00:03:59,000 --> 00:04:00,000 So just save it. 63 00:04:01,000 --> 00:04:05,000 And now just rerun that training cell and book detection. 64 00:04:05,000 --> 00:04:07,000 Just rename this folder. 65 00:04:08,000 --> 00:04:09,000 Copy and just. 66 00:04:10,000 --> 00:04:16,000 Rename this folder and let's train our model on 70 epochs with image size, which we have set in. 67 00:04:16,000 --> 00:04:23,000 The robot was 640, so we will have the same image size and pen book detection data dot yaml file. 68 00:04:23,000 --> 00:04:23,000 Okay. 69 00:04:23,000 --> 00:04:24,000 Looks good. 70 00:04:24,000 --> 00:04:25,000 So just run this. 71 00:04:25,000 --> 00:04:28,000 So if there is an issue we will definitely fix it. 72 00:04:28,000 --> 00:04:30,000 So it might take some time. 73 00:04:30,000 --> 00:04:34,000 So let's see if this is working fine or if there is an issue. 74 00:04:34,000 --> 00:04:35,000 So we can fix it. 75 00:04:35,000 --> 00:04:37,000 So the training is about to start. 76 00:04:37,000 --> 00:04:39,000 So this training process will have an idea. 77 00:04:39,000 --> 00:04:43,000 It will take around one hour, 15 minutes or 2 hours. 78 00:04:43,000 --> 00:04:48,000 So I will stop the video over here as the training begins and as the training completes, I will be 79 00:04:48,000 --> 00:04:48,000 back. 80 00:04:48,000 --> 00:04:54,000 Then I will explain you the whole training process and how we can validate our model and validation 81 00:04:54,000 --> 00:04:59,000 prediction batches, results and how to test our model on the test images. 82 00:04:59,000 --> 00:05:02,000 Plus we will also see how we can test this model on the webcam as well. 83 00:05:03,000 --> 00:05:06,000 So the training has started, but it will take some time. 84 00:05:06,000 --> 00:05:08,000 So as it completes, I will be back. 85 00:05:08,000 --> 00:05:10,000 So see you as the training complete. 86 00:05:13,000 --> 00:05:15,000 Guys the training has completed. 87 00:05:15,000 --> 00:05:18,000 So it took around 39 minutes for the training to complete. 88 00:05:18,000 --> 00:05:25,000 We have trained our model on 70 epochs and we have the last epoch weights while on the 70th epoch and 89 00:05:25,000 --> 00:05:30,000 the best weights while on which our model gives us the best performance or results. 90 00:05:31,000 --> 00:05:33,000 Okay, so here is the model summary. 91 00:05:33,000 --> 00:05:37,000 Or you can say that the overall performance analysis of the model. 92 00:05:37,000 --> 00:05:47,000 So we have got a mean average precision with IOU 50 as on book is 0.295, which is 29.5%. 93 00:05:47,000 --> 00:05:51,000 On the pen, it is 0.157, which is 15.7%. 94 00:05:51,000 --> 00:06:00,000 While the mean average precision when IOU varies from 50 to 95 is with on book is 60.16 8%, which is 95 00:06:00,000 --> 00:06:02,000 16.8%. 96 00:06:02,000 --> 00:06:09,000 And with pen it is 0.0623, which is a bit low, like it's around 6%. 97 00:06:09,000 --> 00:06:12,000 Okay, so let's see, what are the files? 98 00:06:12,000 --> 00:06:14,000 Do we get in the training folder? 99 00:06:14,000 --> 00:06:20,000 So we have the confusion matrix, we have the F1 curve, we have the precision recall curve and we have 100 00:06:20,000 --> 00:06:27,000 the results dot CSV file, which gives us the performance metrics and the performance analysis or metrics 101 00:06:27,000 --> 00:06:29,000 of the model After every epoch. 102 00:06:29,000 --> 00:06:35,000 The result DOT PNG contains the training and validation losses, while the training batch zero gives 103 00:06:35,000 --> 00:06:37,000 us the model prediction on the training batch. 104 00:06:37,000 --> 00:06:44,000 While the validation batch prediction give us the model predictions on the validation pads. 105 00:06:44,000 --> 00:06:50,000 So let us display the confusion matrix and in the weights folder we have the best and large weights 106 00:06:50,000 --> 00:06:51,000 file. 107 00:06:51,000 --> 00:06:56,000 So let us first display the confusion matrix and see what do we get. 108 00:06:56,000 --> 00:07:02,000 So Confusion Matrix is basically a chart that shows how our model handles different classes. 109 00:07:02,000 --> 00:07:09,000 So like you can see that 33% of the times our model detected correctly that there is a book, while 110 00:07:09,000 --> 00:07:16,000 67%, 66% of the times when there is a book, our model was not able to detect that there is a book. 111 00:07:16,000 --> 00:07:21,000 So our model detected nothing 66% of the time while there was a book. 112 00:07:21,000 --> 00:07:25,000 So only 33% of the times our model detected correctly that there is a book. 113 00:07:25,000 --> 00:07:26,000 Why? 114 00:07:27,000 --> 00:07:35,000 11% of the times our model detected correctly that there is a pan one 1% of the time there was a pan. 115 00:07:35,000 --> 00:07:41,000 But our model detected that this is a book, while 88% of the times when there is a pan, our model 116 00:07:41,000 --> 00:07:44,000 was unable to detect anything. 117 00:07:44,000 --> 00:07:49,000 So here we have the training and validation process so you can see that the loss is decreasing a bit. 118 00:07:49,000 --> 00:07:57,000 So if we train our model on a more number of epochs, like 200, 300, we will can see that the we have 119 00:07:57,000 --> 00:08:00,000 the loss will be decreasing and our accuracy will be improving. 120 00:08:00,000 --> 00:08:06,000 Like you can see that mean average precision is increasing like you can see over here and you can see 121 00:08:06,000 --> 00:08:10,000 over here as well, like mean average precision is increasing plus. 122 00:08:11,000 --> 00:08:14,000 Here we can see the model prediction on the validation batch. 123 00:08:14,000 --> 00:08:19,000 So these images are not used for the training, so it's always better to take a look. 124 00:08:19,000 --> 00:08:25,000 Our model is performing on the validation batches, so now we are validating our custom model on the 125 00:08:25,000 --> 00:08:26,000 test dataset. 126 00:08:26,000 --> 00:08:32,000 So the test dataset is not used in the training, so we can validate our model on the custom test dataset. 127 00:08:32,000 --> 00:08:36,000 So here we have validated our model on the test dataset. 128 00:08:36,000 --> 00:08:39,000 So just one thing we have changed in the training process. 129 00:08:39,000 --> 00:08:45,000 We choose mode as train while in the validation, we choose mode as validation and validation. 130 00:08:45,000 --> 00:08:48,000 So here is the other zones of the model. 131 00:08:48,000 --> 00:08:55,000 When we test our model on on the test dataset images, we date our model on the test data set images. 132 00:08:55,000 --> 00:09:06,000 So here we have a mean average precision of 0.29 5.9% .9. 5% of the book and mean average precision 133 00:09:06,000 --> 00:09:09,000 of with 50 port pen is 15.7%. 134 00:09:11,000 --> 00:09:15,000 So here we are testing our model on the test dataset images. 135 00:09:15,000 --> 00:09:18,000 So let's see what results do we get over here. 136 00:09:18,000 --> 00:09:21,000 So here model is unable to predict anything here. 137 00:09:21,000 --> 00:09:25,000 The model has predicted correctly that there is a pen while here. 138 00:09:25,000 --> 00:09:28,000 Our model has also predicted correctly that there is a pen here. 139 00:09:28,000 --> 00:09:34,000 Also, our model predicts correctly that they are walking the shelves and here also on model is predicting 140 00:09:34,000 --> 00:09:35,000 that there are book in the shelves. 141 00:09:35,000 --> 00:09:43,000 So I will download this model weights at best speed and we will test our model on the live webcam as 142 00:09:43,000 --> 00:09:43,000 well. 143 00:09:43,000 --> 00:09:48,000 So test let's test our model on the live webcam and see what results do we get over there.