1 00:00:03,250 --> 00:00:10,480 In this video tutorial, we will learn how we can do object detection using YOLO v ten in Google Colab. 2 00:00:11,140 --> 00:00:16,030 YOLO ten is developed by researchers at Tsinghua University in China. 3 00:00:16,450 --> 00:00:23,170 YOLO ten is the state of the art, real time object detection model that outperforms all the other YOLO 4 00:00:23,170 --> 00:00:29,140 models in terms of average precision, parameter efficiency, and inference speed. 5 00:00:29,470 --> 00:00:35,830 So in this tutorial, we will look at how we can do object detection using YOLO written in Google Colab. 6 00:00:35,860 --> 00:00:39,850 So this is the official repository which you can see over here. 7 00:00:40,120 --> 00:00:44,830 Uh you can find all the graphs over here like average precision and latency. 8 00:00:44,980 --> 00:00:50,470 Uh, comparison latency or accuracy comparison graph where you can see that this is for the Yolo v ten. 9 00:00:50,470 --> 00:00:56,380 And you can see the YOLO v ten outperforms all the other Europe models in terms of accuracy as well 10 00:00:56,380 --> 00:00:58,030 as, uh, in latency. 11 00:00:58,030 --> 00:01:02,710 Like it takes less time as compared to, uh, other YOLO models. 12 00:01:02,710 --> 00:01:04,060 So what is basically latency? 13 00:01:04,060 --> 00:01:09,040 Latency is basically a time taken to do object detection on an input image. 14 00:01:10,250 --> 00:01:10,700 Okay. 15 00:01:10,730 --> 00:01:17,150 So, like, you can see that your lovely tan, uh, takes less inference time as compared to other YOLO 16 00:01:17,150 --> 00:01:21,890 model and in, uh, comparison to other YOLO models like you can see over here. 17 00:01:21,920 --> 00:01:23,990 Uh, these are the number of parameters in billions. 18 00:01:23,990 --> 00:01:29,450 So you can see that, uh, you know, over ten, uh, uh, basically, uh, the architecture of YOLO 19 00:01:29,570 --> 00:01:31,430 ten reduces the computational overhead. 20 00:01:31,430 --> 00:01:36,470 And you can see that the, uh, they are less number of parameters involved in YOLO v ten as compared 21 00:01:36,470 --> 00:01:37,970 to other YOLO models. 22 00:01:37,970 --> 00:01:43,700 And in terms of accuracy, we can clearly see that the model ten outperforms all the other YOLO models. 23 00:01:43,760 --> 00:01:44,000 Okay. 24 00:01:44,150 --> 00:01:50,480 And I can see that ten comes with, uh, six different models Yolov3 ten, nano, small, medium, large, 25 00:01:50,480 --> 00:01:52,100 extra large, all this stuff. 26 00:01:52,100 --> 00:01:57,170 And here you can see all the scripts are written, how we can do object detection using it and how we 27 00:01:57,170 --> 00:02:02,450 can export the model Onnx format and how we can, uh, train the model on any custom data set and how 28 00:02:02,450 --> 00:02:03,650 we can do the predictions. 29 00:02:03,650 --> 00:02:08,030 And here you can find the hugging face, all these things as well. 30 00:02:08,030 --> 00:02:09,950 And here is the citation. 31 00:02:10,250 --> 00:02:12,380 Uh, the code base is built with Ultralytics. 32 00:02:12,380 --> 00:02:14,180 So they have all provided all the details. 33 00:02:14,180 --> 00:02:16,370 And you can check out these details okay. 34 00:02:16,370 --> 00:02:20,120 And you can simply search it on GitHub and you will find this repository. 35 00:02:20,120 --> 00:02:24,530 And if you just click over here you will find the option paper as well. 36 00:02:24,530 --> 00:02:25,940 And you can review this paper. 37 00:02:25,940 --> 00:02:28,610 And I have already made a video tutorial as well. 38 00:02:28,610 --> 00:02:30,920 And in which I have explained the crux of this paper. 39 00:02:30,920 --> 00:02:33,710 Like what approach uh, approach they have adopt. 40 00:02:33,710 --> 00:02:34,790 Like uh. 41 00:02:35,620 --> 00:02:40,150 We all love it and follow the consistent rule assignment strategy. 42 00:02:40,150 --> 00:02:44,200 Plus they also follow efficient driven design strategy as well. 43 00:02:44,200 --> 00:02:46,360 So I've already explained all these things. 44 00:02:46,360 --> 00:02:53,500 What the normal approach they adopt uh, so that they can boost their accuracy and also uh, decrease 45 00:02:53,500 --> 00:02:56,470 the latency as well in France, latency as well. 46 00:02:56,920 --> 00:02:59,800 So I have already explained all these details in other video. 47 00:02:59,800 --> 00:03:03,070 You can check, uh, the video introductions to Eurovision. 48 00:03:04,690 --> 00:03:07,720 So now in the step number one, we need to clone this GitHub repo. 49 00:03:07,720 --> 00:03:11,590 So you can simply copy this link over here and add up this link over here. 50 00:03:11,590 --> 00:03:15,670 And we will just write git clone and set the clone folder as your current directory. 51 00:03:15,700 --> 00:03:21,040 CD stands for current directory, but please make sure that you have selected the runtime as your GPU 52 00:03:21,040 --> 00:03:22,390 so we are good to go ahead. 53 00:03:23,900 --> 00:03:26,300 So first of all, we'll clone this repository over here. 54 00:03:26,630 --> 00:03:27,050 Uh, okay. 55 00:03:27,050 --> 00:03:29,510 So now you can see we have the clone repository over here. 56 00:03:29,510 --> 00:03:31,970 And now we will install all the required packages. 57 00:03:33,670 --> 00:03:39,700 So now all the required packages will be installed that are required to do object detection on an image, 58 00:03:39,700 --> 00:03:41,410 video or a live webcam feed. 59 00:03:41,410 --> 00:03:43,840 So this will install all the required packages over here. 60 00:03:44,260 --> 00:03:48,460 So this will take few seconds before all the required packages are get installed. 61 00:03:52,080 --> 00:03:53,640 So meanwhile, it gets installed. 62 00:03:53,670 --> 00:03:56,220 Let me show you, uh, the pre-trained model words. 63 00:03:56,220 --> 00:03:57,180 Where are they available? 64 00:03:57,180 --> 00:04:02,010 So if you just click over here, uh, you can find all the pre-trained models weights over here. 65 00:04:02,010 --> 00:04:08,610 Your written small user written nano zero V ten x large and medium and user written large model here. 66 00:04:08,610 --> 00:04:12,210 So you can simply copy the link address over here okay. 67 00:04:12,540 --> 00:04:19,920 So and if you just paste this link address over here uh like simply if you just place this link address 68 00:04:19,920 --> 00:04:24,330 over here, you will find the link to your written nano model weights okay. 69 00:04:25,080 --> 00:04:31,080 So simply you just need to click over here and need to copy the link address. 70 00:04:31,080 --> 00:04:32,880 And you just need to add a link address. 71 00:04:32,880 --> 00:04:35,610 All these model weights link address over here. 72 00:04:35,610 --> 00:04:37,980 And uh I have already done the script. 73 00:04:37,980 --> 00:04:43,800 It will download the pre-trained model weights uh over here into this YOLO written repository. 74 00:04:43,800 --> 00:04:46,950 So in currently all the required packages that get installed. 75 00:04:47,040 --> 00:04:50,520 Uh, so this will take few more, uh, seconds before they get installed. 76 00:04:53,880 --> 00:04:55,920 So now you can see all the packages that are installed. 77 00:04:55,920 --> 00:04:58,650 So now we will download the model weights over there. 78 00:04:58,650 --> 00:05:00,180 So you can see we run this cell. 79 00:05:01,860 --> 00:05:02,130 And all. 80 00:05:02,130 --> 00:05:06,000 You can see that, um, all the models will get downloaded over here, like in. 81 00:05:06,000 --> 00:05:06,450 Fine. 82 00:05:07,080 --> 00:05:07,890 Okay. 83 00:05:07,920 --> 00:05:10,350 Let us get some leads and I will show you. 84 00:05:11,190 --> 00:05:14,850 So now you can see how weird the directory is being created over here. 85 00:05:14,850 --> 00:05:18,720 And we have added all the model weights over here like in find over here. 86 00:05:19,170 --> 00:05:19,530 Okay. 87 00:05:19,530 --> 00:05:22,650 So now we will be doing inference for an image using the pre-trained model. 88 00:05:22,650 --> 00:05:26,580 I have added, uploaded and input image on my Google drive. 89 00:05:26,580 --> 00:05:31,590 So I have just added the link over here and I'm directly downloading the input image from drive into 90 00:05:31,590 --> 00:05:33,360 this Google Colab notebook over here. 91 00:05:36,330 --> 00:05:36,870 Well, good. 92 00:05:37,620 --> 00:05:37,890 Okay. 93 00:05:37,890 --> 00:05:39,660 So now we can see we have the image. 94 00:05:39,660 --> 00:05:43,500 So we will doing object detection on this input image. 95 00:05:43,650 --> 00:05:44,280 Okay. 96 00:05:44,610 --> 00:05:51,510 So now you can see we have the bus traffic lights persons bicycle and all the stuff over here. 97 00:05:51,510 --> 00:05:54,150 Like you can see we have the first pathway as well. 98 00:05:54,150 --> 00:05:57,210 So let's do object detection on this input image. 99 00:05:57,210 --> 00:06:00,540 So we you just sitting right over here cost is equal to detect. 100 00:06:00,540 --> 00:06:04,350 We are doing prediction and we want to save the output image. 101 00:06:04,350 --> 00:06:08,520 And here you will pass the model weights over here you can use any other model weights over here. 102 00:06:08,520 --> 00:06:12,630 Like you can find all different model weights in which we can comes with. 103 00:06:12,660 --> 00:06:14,820 You can use any other model weights as well. 104 00:06:14,820 --> 00:06:17,400 And here you just pass the name of the input image. 105 00:06:17,610 --> 00:06:19,710 You are doing object detection on video. 106 00:06:19,710 --> 00:06:22,530 You will pass the video name over here okay. 107 00:06:22,530 --> 00:06:23,460 And let's see. 108 00:06:27,230 --> 00:06:28,310 Now how it works. 109 00:06:32,820 --> 00:06:38,310 But now you can see we have done detections of one bicycle, 12 persons, one train, three traffic 110 00:06:38,310 --> 00:06:45,390 lights, two bag bags, four handbags are detected and the inference time is taken as 95.5 millisecond, 111 00:06:45,390 --> 00:06:48,090 the time taken to do prediction on this input image. 112 00:06:48,840 --> 00:06:51,120 So let me display this output image over here. 113 00:06:51,600 --> 00:06:53,880 So here we have the output image. 114 00:06:53,880 --> 00:06:57,180 But you can see over here bicycle train traffic lights. 115 00:06:57,180 --> 00:07:01,230 This person which are very far away are also detected with a very good confidence score as well. 116 00:07:01,230 --> 00:07:03,720 Like can see over here and bag backpack. 117 00:07:03,720 --> 00:07:06,510 All this stuff is also detected and it looks good. 118 00:07:07,080 --> 00:07:10,980 Uh, now we will doing a friends on a video using the pre-trained model. 119 00:07:11,220 --> 00:07:13,380 Okay, so let's do this. 120 00:07:13,380 --> 00:07:18,600 I'm downloading a sample video from worldwide directly to this Google Colab notebook over here. 121 00:07:23,660 --> 00:07:27,710 Okay, so let's see how it goes. 122 00:07:33,690 --> 00:07:39,000 So now you can see, uh, all the complete video is being divided into 1314 frames. 123 00:07:39,000 --> 00:07:42,120 And we are doing object detection on each of the frames as well. 124 00:07:42,120 --> 00:07:46,410 And then in the output we get a complete video output as well. 125 00:07:46,440 --> 00:07:46,770 Okay. 126 00:07:46,770 --> 00:07:51,420 So now it will take few seconds to do object detection on this complete video. 127 00:07:56,170 --> 00:07:58,870 The object detection on this input video is being done. 128 00:07:58,870 --> 00:08:02,140 Like you can see for all the 1340 frames. 129 00:08:02,140 --> 00:08:02,710 It's done. 130 00:08:02,710 --> 00:08:05,020 And here we have the output video. 131 00:08:05,110 --> 00:08:09,340 So let me download this output video and show you on the screen how does it looks like? 132 00:08:10,360 --> 00:08:10,630 Uh. 133 00:08:22,310 --> 00:08:22,760 Us. 134 00:08:22,880 --> 00:08:25,250 Well, now you can see over here. 135 00:08:25,430 --> 00:08:31,340 Uh, here we have the output video, but like you can see over here, um, we are able to detect the 136 00:08:31,340 --> 00:08:32,810 cars over here. 137 00:08:32,900 --> 00:08:33,410 Okay. 138 00:08:33,410 --> 00:08:35,990 And the truck over here is also detected. 139 00:08:35,990 --> 00:08:41,570 We have detected, uh, this is a false positive like was shown previously. 140 00:08:41,870 --> 00:08:44,270 Um, but you can see the detection results are good. 141 00:08:44,270 --> 00:08:46,220 Like, we are able to detect the cars truck. 142 00:08:46,280 --> 00:08:52,040 So now you can see over here we are able to do object detection on an image and, and as well as on 143 00:08:52,040 --> 00:08:53,210 the video as well. 144 00:08:53,210 --> 00:08:56,900 And we have written gives quite good results as well. 145 00:08:57,230 --> 00:08:59,390 So that's all from this tutorial. 146 00:08:59,390 --> 00:09:05,960 In this tutorial we have seen that how we can do object detection on an image and on video using uh, 147 00:09:05,960 --> 00:09:06,980 YOLO written. 148 00:09:07,160 --> 00:09:07,640 Okay. 149 00:09:07,640 --> 00:09:08,690 Thank you for watching.