1 00:00:02,940 --> 00:00:10,860 In this video tutorial, I will be testing the Yolo v9 model performance on image video and on the live 2 00:00:10,860 --> 00:00:11,970 webcam feed. 3 00:00:11,970 --> 00:00:14,250 So let's get started with it. 4 00:00:14,700 --> 00:00:18,990 I have divided this complete tutorial into six different steps. 5 00:00:18,990 --> 00:00:22,530 In the step number one, I will go on the yolo v9 GitHub repo. 6 00:00:23,160 --> 00:00:28,260 In the step number two, I will create a virtual environment which is recommended if you don't want 7 00:00:28,260 --> 00:00:30,120 to disturb your Python packages. 8 00:00:30,600 --> 00:00:35,490 Then after creating a virtual environment in the step number three, I will install all the required 9 00:00:35,490 --> 00:00:41,970 packages or libraries that are required to do object detection or image video and on the live webcam 10 00:00:41,970 --> 00:00:42,480 feed. 11 00:00:42,870 --> 00:00:46,530 In the step number four, I will do object detection on image. 12 00:00:46,530 --> 00:00:52,350 I will be showing, showing you how you can play with different hyper parameters and adjust them as 13 00:00:52,350 --> 00:00:53,850 per your requirements. 14 00:00:53,850 --> 00:00:57,480 Then in the step number five, we will be doing object detection on video. 15 00:00:57,480 --> 00:01:02,310 And in the last step I will show you how you can do object detection on the live webcam feed. 16 00:01:02,310 --> 00:01:04,500 So let's get started with it. 17 00:01:05,010 --> 00:01:10,650 So over here you can see that I have created a project by the name testing YOLO Benign Model Performance. 18 00:01:10,650 --> 00:01:12,390 In my local app directory. 19 00:01:12,390 --> 00:01:17,100 You can create a project in any of your directory, and I am using PyCharm Community Edition. 20 00:01:17,100 --> 00:01:20,190 And I have Python 3.10 installed. 21 00:01:20,520 --> 00:01:21,030 Okay. 22 00:01:22,990 --> 00:01:23,170 Going. 23 00:01:23,170 --> 00:01:26,020 The step number one, we need to clone the Yolov5 and GitHub repo. 24 00:01:26,020 --> 00:01:28,000 So I will just go over here. 25 00:01:30,030 --> 00:01:37,290 Yolo v9 GitHub and I will just open the first link from here, and just go over here and just copy the 26 00:01:37,290 --> 00:01:40,470 URL to the clipboard, and then go back to the PyCharm. 27 00:01:40,470 --> 00:01:48,810 And over here I will just write let clone and just uh, pass the link or copy the URL link, which I 28 00:01:48,810 --> 00:01:49,320 pass. 29 00:01:49,560 --> 00:01:51,420 Uh, just paste the link over here. 30 00:01:51,540 --> 00:01:54,600 So now I will clone the complete Golovnin GitHub repo. 31 00:01:59,050 --> 00:02:02,230 So now in the next step, I will just go to the clone folder. 32 00:02:02,230 --> 00:02:04,450 So if I just, uh, refresh. 33 00:02:04,450 --> 00:02:04,690 Okay. 34 00:02:04,690 --> 00:02:06,010 No need, no need to refresh. 35 00:02:06,010 --> 00:02:08,080 We can find this over here. 36 00:02:08,080 --> 00:02:12,790 So we have all the benign repository being cloned over here. 37 00:02:12,790 --> 00:02:15,850 And we have all the required files we can see over here. 38 00:02:16,120 --> 00:02:16,420 Okay. 39 00:02:16,420 --> 00:02:16,780 So. 40 00:02:17,370 --> 00:02:20,280 So in the next step I will just go to the clone folder. 41 00:02:20,280 --> 00:02:23,550 So I will just set CD current directory yolo v nine. 42 00:02:23,550 --> 00:02:25,800 So now we are going to the clone folder. 43 00:02:26,190 --> 00:02:28,200 So now we are here at this clone folder. 44 00:02:28,200 --> 00:02:28,470 Now. 45 00:02:30,670 --> 00:02:33,370 So now we have done with the step number one. 46 00:02:33,370 --> 00:02:36,310 Then the step number two will create a virtual environment. 47 00:02:36,790 --> 00:02:37,180 Okay. 48 00:02:37,180 --> 00:02:40,480 So to create a virtual environment I will write over here. 49 00:02:42,500 --> 00:02:45,140 I on, I am. 50 00:02:48,340 --> 00:02:53,290 Virtual environment, and I will be creating a virtual environment by the name YOLO v9 testing. 51 00:02:53,290 --> 00:02:55,750 So this will be the name of my virtual environment. 52 00:02:55,780 --> 00:02:56,980 Yolo v9 testing. 53 00:03:02,430 --> 00:03:04,110 This will take few more seconds. 54 00:03:12,490 --> 00:03:12,790 It will. 55 00:03:12,790 --> 00:03:14,830 Let's wait for it to get completed. 56 00:03:14,830 --> 00:03:16,000 Okay, so it's done. 57 00:03:16,540 --> 00:03:17,770 So if I just. 58 00:03:18,880 --> 00:03:19,540 Your ballot. 59 00:03:19,540 --> 00:03:22,150 Here we have this jewelry nine. 60 00:03:24,040 --> 00:03:25,630 Let's think forward here. 61 00:03:27,820 --> 00:03:28,210 Benign. 62 00:03:28,210 --> 00:03:30,730 The next step we will go to this scripts directory. 63 00:03:30,730 --> 00:03:31,960 So I will just write. 64 00:03:34,450 --> 00:03:37,000 First we need to go to the YOLO v9 testing. 65 00:03:37,000 --> 00:03:40,330 So I will just write YOLO in testing folder. 66 00:03:41,360 --> 00:03:41,750 Okay. 67 00:03:41,750 --> 00:03:44,300 And then we need to go to the scripts directory. 68 00:03:44,300 --> 00:03:47,480 So I will just write CD scripts. 69 00:03:50,040 --> 00:03:51,990 And I will just activate this. 70 00:03:54,080 --> 00:03:56,150 Okay, so it's now activated. 71 00:03:56,900 --> 00:03:59,000 So now I will just go and. 72 00:04:03,390 --> 00:04:04,290 But that's done. 73 00:04:04,290 --> 00:04:09,390 So now in the next step, what we'll do is we will, uh, install all the required packages that we 74 00:04:09,390 --> 00:04:12,120 have in the requirements.txt file. 75 00:04:12,120 --> 00:04:18,600 So if I open the requirements.txt file, we can see all the required packages that are required to do 76 00:04:18,600 --> 00:04:22,950 object detection on image video and on the live webcam feed using YOLO. 77 00:04:22,950 --> 00:04:24,930 Benign are being listed over here. 78 00:04:25,140 --> 00:04:32,880 To install all these required packages we just need to write pip install minus r requirements dot txt 79 00:04:32,880 --> 00:04:34,530 and let's enter. 80 00:04:48,050 --> 00:04:51,290 So this package installation will take time. 81 00:04:51,290 --> 00:04:55,430 So if, uh, you can see over here I have written minus r. 82 00:04:55,430 --> 00:04:58,460 So this is basically a hyphen RI1. 83 00:04:58,460 --> 00:04:59,900 You can say that hyphen r. 84 00:04:59,900 --> 00:05:07,760 So basically it means that uh recursively for each library will be installed that are listed in this 85 00:05:07,820 --> 00:05:09,680 requirements.txt file. 86 00:05:10,190 --> 00:05:13,400 So this package installation will take some time. 87 00:05:13,400 --> 00:05:15,230 So let's wait for it to get completed. 88 00:05:19,670 --> 00:05:23,090 Now you can see that all the required packages are being installed. 89 00:05:23,090 --> 00:05:25,850 So now we will do object detection on image. 90 00:05:26,090 --> 00:05:29,540 So I will just go to the GitHub repo over here. 91 00:05:30,370 --> 00:05:32,770 So if I just go down over here. 92 00:05:32,770 --> 00:05:35,260 So now we will be using the following nine models. 93 00:05:35,260 --> 00:05:41,890 And to do object detection on image using following nine model we will be using the detect.py script 94 00:05:41,890 --> 00:05:42,340 file. 95 00:05:42,430 --> 00:05:46,510 And then we will use this command okay. 96 00:05:46,930 --> 00:05:49,630 So let's move forward here. 97 00:05:49,630 --> 00:05:51,370 So detect-2.py. 98 00:05:51,370 --> 00:05:53,860 So let's see how we can do object detection on image. 99 00:05:53,860 --> 00:05:57,670 So I will be using Python detect dash dot py. 100 00:05:58,090 --> 00:06:02,230 And I have just created a folder by the name over here resources. 101 00:06:02,290 --> 00:06:05,800 So here I've added three different images okay. 102 00:06:05,800 --> 00:06:08,110 So now we will be doing object detection on these images. 103 00:06:08,110 --> 00:06:13,630 And I have also added two videos over here as well okay so that's fine. 104 00:06:13,900 --> 00:06:16,690 So then I will just define. 105 00:06:17,710 --> 00:06:19,840 Thought over here as. 106 00:06:22,120 --> 00:06:25,270 We'll go to the resource space for the. 107 00:06:29,140 --> 00:06:33,700 And then I will be doing object detection on image one dot jpg first. 108 00:06:34,390 --> 00:06:37,480 Okay, so I just need to pass the file name over here. 109 00:06:37,480 --> 00:06:38,440 The spelling of. 110 00:06:39,510 --> 00:06:40,920 Beaux Arts. 111 00:06:40,920 --> 00:06:41,280 Correct. 112 00:06:42,090 --> 00:06:45,210 So let me add weights over here as well. 113 00:06:45,210 --> 00:06:47,730 So if you just go over here. 114 00:06:48,210 --> 00:06:53,340 So and if you just if over here so go releases over here. 115 00:06:53,730 --> 00:06:57,240 So now you can see that you'll be nine comes with four different models. 116 00:06:57,240 --> 00:07:03,540 But currently we have uh nine weights available for two models which is YOLO v9 compact and YOLO v9 117 00:07:03,540 --> 00:07:04,140 extended. 118 00:07:04,290 --> 00:07:07,950 So I will be downloading the YOLO v9 compact model weights. 119 00:07:07,950 --> 00:07:11,880 And we will be doing testing using the YOLO v9 compact model weights. 120 00:07:11,910 --> 00:07:15,120 So now you can see that V9 comes with different models. 121 00:07:15,120 --> 00:07:20,310 But currently the weights of YOLO one and compact model and the V9 extended model are available. 122 00:07:20,310 --> 00:07:23,040 And we will be using the YOLO v9 compact model weights. 123 00:07:23,040 --> 00:07:29,550 Although YOLO v9 extended model weight is better, gives better accuracy than YOLO, and compact model 124 00:07:29,550 --> 00:07:33,480 like YOLO v nine extended model outperforms uh income. 125 00:07:33,480 --> 00:07:36,900 Uh, all the other YOLO v9 models like you can see over here. 126 00:07:39,210 --> 00:07:46,020 Well, now you can see the file is downloaded and I will just go over here, go under downloads. 127 00:07:46,650 --> 00:07:49,380 Copy this from here and go into my. 128 00:07:50,680 --> 00:07:55,300 Wandered over here and I'll just play it again. 129 00:07:59,710 --> 00:08:03,040 And so there we have that model. 130 00:08:03,340 --> 00:08:05,110 And that's all you need to do. 131 00:08:06,820 --> 00:08:07,240 Okay. 132 00:08:07,240 --> 00:08:09,310 So let's go backward here. 133 00:08:09,550 --> 00:08:14,170 And if you just see over here we have the model weights available over here. 134 00:08:14,530 --> 00:08:17,860 And if I just go below down over here. 135 00:08:25,450 --> 00:08:26,110 Okay. 136 00:08:26,830 --> 00:08:29,260 So I will just passing the normal weights. 137 00:08:29,260 --> 00:08:32,830 So first the model weights I will be writing weights. 138 00:08:34,780 --> 00:08:38,260 Then we will just go to the weights rectory. 139 00:08:38,260 --> 00:08:41,770 And over here we have the following line. 140 00:08:42,600 --> 00:08:43,200 Bad. 141 00:08:43,200 --> 00:08:47,670 Seen it all and let's see how it goes. 142 00:08:49,230 --> 00:08:51,090 So there are some stay over there. 143 00:08:56,540 --> 00:08:59,180 Here is our Lord and let's see how it works now. 144 00:09:14,780 --> 00:09:16,760 But this will take two more seconds. 145 00:09:18,020 --> 00:09:23,810 I don't have a GPU available so I'm doing are testing it on my CPU. 146 00:09:23,810 --> 00:09:24,920 Okay now. 147 00:09:25,070 --> 00:09:25,580 So here. 148 00:09:29,420 --> 00:09:32,510 Opening, it will work fine, but let's see. 149 00:09:45,040 --> 00:09:45,430 Okay. 150 00:09:45,430 --> 00:09:50,020 So now we have the output into this runs segmentation do image one. 151 00:09:50,020 --> 00:09:54,520 So this is our output image which you can see over here. 152 00:09:55,090 --> 00:09:55,570 Okay. 153 00:09:56,140 --> 00:10:04,150 So now you can see over here we are able to track the person dog okay. 154 00:10:04,150 --> 00:10:10,600 And here you can see that uh the car is quite blurred but our model is available uh also able to detect 155 00:10:10,600 --> 00:10:11,890 that car as well. 156 00:10:12,250 --> 00:10:14,500 But, uh, this is a wrong detection. 157 00:10:14,500 --> 00:10:16,210 This is complete backpack. 158 00:10:17,110 --> 00:10:20,830 Okay, so now one wrong direction is look fine. 159 00:10:21,430 --> 00:10:22,000 Okay. 160 00:10:22,000 --> 00:10:30,580 So, for example, if, uh, you got a case where you have very small objects, okay. 161 00:10:30,790 --> 00:10:36,520 And you have to decrease down the confidence value, for example, currently my confidence threshold 162 00:10:36,520 --> 00:10:38,770 is being set as 0.25. 163 00:10:39,310 --> 00:10:43,120 Like you can see over here, the default confidence value is 0.25. 164 00:10:43,600 --> 00:10:45,940 So what is this confidence value means. 165 00:10:46,030 --> 00:10:46,870 So. 166 00:10:47,510 --> 00:10:54,830 After doing uh, so object detection using yolo v9, we will only be drawing the bounding boxes around 167 00:10:54,830 --> 00:10:55,220 that. 168 00:10:55,220 --> 00:11:01,880 Around around that, objects that are on that, our confidence threshold or confidence value above 25%. 169 00:11:02,030 --> 00:11:02,360 Okay. 170 00:11:02,360 --> 00:11:03,710 So this is the confidence threshold. 171 00:11:03,710 --> 00:11:11,240 So all the detected objects that have a confidence uh below 25%, uh, we will not be drawing bounding 172 00:11:11,240 --> 00:11:12,140 box around them. 173 00:11:12,530 --> 00:11:12,860 Okay. 174 00:11:12,860 --> 00:11:18,530 So now you can see here the confidence value is 0.900.780.. 175 00:11:18,530 --> 00:11:20,120 Like something here as well. 176 00:11:20,120 --> 00:11:22,130 Like it's not visible over here. 177 00:11:22,130 --> 00:11:26,720 But we can see currently the back 0.48 the all the uh. 178 00:11:28,050 --> 00:11:33,960 All the objects that have bounding boxes drawn around them have a confidence value above 25%. 179 00:11:33,960 --> 00:11:34,380 Okay. 180 00:11:34,740 --> 00:11:39,390 So like we are only drawing bounding boxes around those objects which have a confidence value above 181 00:11:39,390 --> 00:11:40,410 25%. 182 00:11:40,410 --> 00:11:45,780 So what this confidence value means, the confidence value means that the model is 25%. 183 00:11:45,930 --> 00:11:51,690 For example, I've said that for for example, the confidence value over here is 0.90 for dog. 184 00:11:51,690 --> 00:11:55,800 So this means that the model is 90% confident that this is a dog. 185 00:11:56,310 --> 00:11:57,030 Here is person. 186 00:11:57,030 --> 00:11:59,610 For the person, the confidence value is 0.78. 187 00:11:59,730 --> 00:12:04,560 This means the model is 78% confident that the this is a person. 188 00:12:04,560 --> 00:12:04,980 Okay. 189 00:12:05,400 --> 00:12:08,910 And like this is what we meant by confidence value. 190 00:12:08,910 --> 00:12:12,330 So and here I have defined the confidence threshold as 0.25. 191 00:12:12,330 --> 00:12:13,350 So this means. 192 00:12:14,000 --> 00:12:19,940 All the objects that have a confidence value, all the detected objects that have a confidence value 193 00:12:19,940 --> 00:12:23,720 above 25%, we will be drawing bounding boxes around them. 194 00:12:23,720 --> 00:12:29,180 Like you can see, we have drawn bounding boxes for a dog, for a person, for a god, and that's it. 195 00:12:29,510 --> 00:12:35,300 Okay, so there might be some case where you have to decrease down this confidence value to do object 196 00:12:35,300 --> 00:12:35,900 detection. 197 00:12:35,900 --> 00:12:36,380 Okay. 198 00:12:36,740 --> 00:12:37,790 For example. 199 00:12:41,560 --> 00:12:50,200 If I decrease the confidence value to 0.10 and let's see, what results do I get from here. 200 00:12:50,200 --> 00:12:54,190 So now by default the confidence is back. 201 00:12:54,850 --> 00:12:58,000 So I just need to write 1% threshold. 202 00:12:59,320 --> 00:13:02,950 So now I have decreased the confidence threshold to 0.10. 203 00:13:02,950 --> 00:13:10,360 So what I mean over here is that the all the objects that were benign model detect all the objects that 204 00:13:10,360 --> 00:13:16,720 my YOLO benign model, that which have confidence value above 10% will have bounding boxes drawn around 205 00:13:16,720 --> 00:13:16,990 them. 206 00:13:16,990 --> 00:13:17,470 Okay. 207 00:13:23,090 --> 00:13:25,790 But now you can see, uh, I don't see any. 208 00:13:27,320 --> 00:13:28,460 Back over here. 209 00:13:30,080 --> 00:13:33,230 Let's decrease the confidence value to 0.05. 210 00:13:33,230 --> 00:13:43,070 Like all the detected objects that have a confidence value above 5% will have bounding boxes drawn around 211 00:13:43,070 --> 00:13:43,310 them. 212 00:14:00,460 --> 00:14:03,400 So now you can see over here okay. 213 00:14:03,400 --> 00:14:06,520 So now you can see we have detected a handbag over here as well. 214 00:14:06,940 --> 00:14:07,300 Okay. 215 00:14:07,300 --> 00:14:12,910 So let's decrease down the confidence value to 0.101%. 216 00:14:12,910 --> 00:14:18,340 So this means that after doing object detection using YOLO benign, all the detected objects that were 217 00:14:18,340 --> 00:14:22,990 confidence value above 1% will have bounding boxes drawn around them. 218 00:14:31,800 --> 00:14:33,990 So if I just, like, crash over here. 219 00:14:35,680 --> 00:14:39,760 So now you can see that this is what I want to show you over here. 220 00:14:40,000 --> 00:14:45,100 So if you just see over here, let me open this image from here as well. 221 00:14:59,140 --> 00:15:01,000 So this is the output image which we have gone to. 222 00:15:01,000 --> 00:15:01,150 Now. 223 00:15:01,150 --> 00:15:04,810 You can see over here, uh this is the backpack okay. 224 00:15:04,810 --> 00:15:06,640 So this is correct direction. 225 00:15:06,640 --> 00:15:11,650 This is the person over here like you can see over here. 226 00:15:11,650 --> 00:15:15,160 We have detected person two times okay. 227 00:15:15,550 --> 00:15:20,860 So now you can see there are two persons like it has detected person 2.42 times. 228 00:15:20,950 --> 00:15:23,470 Although there is only one person okay. 229 00:15:23,470 --> 00:15:28,870 Like you can see over here there are two person like it has detected person two times but there is only 230 00:15:28,870 --> 00:15:29,830 one person. 231 00:15:29,980 --> 00:15:31,870 Okay so. 232 00:15:31,870 --> 00:15:38,230 But last you can see over here, uh, we have uh, detected two backpacks like you can see one over 233 00:15:38,230 --> 00:15:39,850 here, one over here. 234 00:15:40,270 --> 00:15:40,600 Okay. 235 00:15:40,600 --> 00:15:44,980 So now you can see it has detected three backpacks, although there is only one backpack. 236 00:15:44,980 --> 00:15:48,010 So like you can see that we have, uh. 237 00:15:49,310 --> 00:15:54,170 Multiple bonding boxes around a single object, like there is one person, but you can see we have two 238 00:15:54,170 --> 00:15:54,950 bonding boxes. 239 00:15:54,950 --> 00:15:59,990 There is one backpack, but you can see we have three bounding boxes around the backpacks. 240 00:15:59,990 --> 00:16:00,410 Okay. 241 00:16:04,020 --> 00:16:04,590 So. 242 00:16:04,590 --> 00:16:10,140 But to remove this overlapping bounding boxes like you can see that for person we have two overlapping 243 00:16:10,140 --> 00:16:10,800 bounding boxes. 244 00:16:10,800 --> 00:16:14,400 For backpack we have three overlapping bounding boxes now. 245 00:16:14,400 --> 00:16:19,410 So to remove this overlapping bounding boxes we use a technique called non-max pressure. 246 00:16:19,410 --> 00:16:25,050 So non-max suppression is a technique that is used to remove, uh, overlapping bounding boxes that 247 00:16:25,050 --> 00:16:27,780 may arise from the object detection algorithm. 248 00:16:28,710 --> 00:16:29,220 Okay. 249 00:16:31,050 --> 00:16:33,390 So what is the main idea of norm expression? 250 00:16:33,390 --> 00:16:39,570 So the main idea for the norm expression is to retain only the bounding boxes that have a high confidence 251 00:16:39,570 --> 00:16:40,860 score than other. 252 00:16:41,190 --> 00:16:41,580 Okay. 253 00:16:41,580 --> 00:16:47,460 So what will happen is that we will be using norm expression to remove this, uh, like multiple bounding 254 00:16:47,460 --> 00:16:50,550 boxes that have been drawn around a single object, like two parts. 255 00:16:50,670 --> 00:16:53,190 We have only one person, but we have two bounding boxes. 256 00:16:53,280 --> 00:16:55,890 Uh, like the model has detected two persons a plus. 257 00:16:55,890 --> 00:16:59,250 You can see we have only one backpack, and the model has detected three backpacks. 258 00:16:59,250 --> 00:16:59,670 Okay. 259 00:16:59,790 --> 00:17:04,140 So now you can see we have multiple bounding boxes, uh, for one single object. 260 00:17:04,140 --> 00:17:10,050 So to remove this overlapping bounding boxes or redundant or extra bounding boxes, we will be using 261 00:17:10,050 --> 00:17:13,620 a technique called norm expression which will retain only uh. 262 00:17:13,620 --> 00:17:18,420 So now you out of this multiple bounding boxes for a single object like person, like we have created 263 00:17:18,420 --> 00:17:23,550 two persons, but we will only retain only one person, and that person will be retained that have a 264 00:17:23,550 --> 00:17:25,440 high confidence value. 265 00:17:25,440 --> 00:17:25,860 Okay. 266 00:17:26,100 --> 00:17:28,290 So what will I will be use for this? 267 00:17:28,290 --> 00:17:31,110 I will be using a technique called norm expression. 268 00:17:31,110 --> 00:17:35,880 And here you can see we have a IOU threshold for the norm expression. 269 00:17:35,880 --> 00:17:39,570 So this IOU is what non-maximum suppression. 270 00:17:43,150 --> 00:17:48,640 But the value of this IOU threshold for the non-maximum suppression varies from 0 to 1. 271 00:17:48,700 --> 00:17:51,070 So this value will be between 0 to 1. 272 00:17:51,650 --> 00:17:55,220 So if we set the higher threshold value, like currently it's in the middle. 273 00:17:55,220 --> 00:18:00,530 But if we set the higher threshold value like 0.80.9, what will happen. 274 00:18:00,560 --> 00:18:02,510 This will result in a. 275 00:18:04,610 --> 00:18:09,320 So what will happen is that the this only lapping bounding boxes will not be removed. 276 00:18:09,320 --> 00:18:14,600 If we set the original threshold for the non expression of value high like it varies from 0 to 1. 277 00:18:14,600 --> 00:18:17,420 And if I set the value 0.8 to 0.9. 278 00:18:17,420 --> 00:18:23,090 So this overlapping bounding boxes will not be removed or very few will be removed. 279 00:18:23,090 --> 00:18:29,510 And all the overlapping bounding boxes or the maximum number of overlapping bounding boxes will be retained. 280 00:18:32,970 --> 00:18:33,300 Okay. 281 00:18:33,300 --> 00:18:38,790 And if I just set the, uh, I use threshold value to low, like 0.1 or 0.2. 282 00:18:38,790 --> 00:18:43,320 So this will result in all the overlapping bounding boxes will be removed. 283 00:18:43,320 --> 00:18:46,950 And this will result in a fewer but more accurate detection. 284 00:18:46,950 --> 00:18:51,600 So if I set up a threshold for non expression as a very low value. 285 00:18:51,600 --> 00:18:55,710 So this will result in uh overlapping bounding boxes will be removed. 286 00:18:55,710 --> 00:18:59,310 And this will result in fewer but more accurate detections. 287 00:19:00,800 --> 00:19:02,510 So let's test this out. 288 00:19:02,510 --> 00:19:07,040 So if I just set out the IOU threshold over here. 289 00:19:12,540 --> 00:19:13,740 0.1. 290 00:19:13,800 --> 00:19:13,980 Okay. 291 00:19:13,980 --> 00:19:17,610 So currently you can see that we have only 1% but 2% collected. 292 00:19:17,610 --> 00:19:20,280 We have only one backpack but three backpacks are detected. 293 00:19:20,640 --> 00:19:24,990 So let's see if I just set this threshold value okay. 294 00:19:24,990 --> 00:19:28,470 And I will stake its IOU ratio. 295 00:19:29,850 --> 00:19:32,670 So let's see if I just set a small value of higher threshold. 296 00:19:32,670 --> 00:19:34,440 Will it affect my result. 297 00:19:34,440 --> 00:19:37,620 Will it remove the overlapping bounding boxes that are being drawn? 298 00:19:47,600 --> 00:19:49,700 So now you can see that only we have one person. 299 00:19:49,700 --> 00:19:51,680 Now we have one backpack. 300 00:19:51,680 --> 00:19:52,340 And okay. 301 00:19:52,340 --> 00:19:56,420 So now you can see that here we have only okay okay. 302 00:19:56,420 --> 00:20:00,710 So now you can see that in the previous case we have two persons and three and backpacks. 303 00:20:00,710 --> 00:20:03,620 But now you can see we have only one person and one backpack. 304 00:20:03,620 --> 00:20:07,610 Like this means on the overlapping bounding boxes have been removed. 305 00:20:07,610 --> 00:20:09,410 And if I just show you over here. 306 00:20:12,090 --> 00:20:16,080 So now, if I just show you this output over here as well. 307 00:20:21,660 --> 00:20:24,450 Well, now you can see over here like. 308 00:20:25,290 --> 00:20:27,420 The is. 309 00:20:30,670 --> 00:20:35,050 So now you can see we have only one bounding box around the person like you can see over here. 310 00:20:35,050 --> 00:20:36,730 We have only one handbag. 311 00:20:36,730 --> 00:20:37,090 Okay. 312 00:20:37,090 --> 00:20:39,310 Over here we have only one dog. 313 00:20:39,310 --> 00:20:42,130 Like all the overlapping bounding boxes have been removed. 314 00:20:42,130 --> 00:20:42,670 Okay. 315 00:20:42,730 --> 00:20:45,820 So now you can see that my output looks very fine. 316 00:20:45,820 --> 00:20:48,370 Like I have not changed the confidence threshold. 317 00:20:48,730 --> 00:20:52,210 I only adjusted the threshold for non expression. 318 00:20:52,210 --> 00:20:55,060 And this has removed all the overlapping bounding boxes. 319 00:20:55,390 --> 00:20:59,350 Like you can see that all the overlapping bounding boxes have been removed. 320 00:20:59,350 --> 00:20:59,770 Okay. 321 00:20:59,800 --> 00:21:07,750 So like you can see I have set it the very small value for the threshold for non expression which you 322 00:21:07,750 --> 00:21:09,190 can see over here. 323 00:21:11,330 --> 00:21:16,820 Okay, I've just set the value 0.1 and this resulted in all the overlapping bounding boxes being removed. 324 00:21:16,820 --> 00:21:22,100 And if I just set the threshold for non expression as a high value like 0.9. 325 00:21:22,100 --> 00:21:25,340 So the overlapping bounding boxes will be retained. 326 00:21:25,340 --> 00:21:29,930 And this will like maximum number of overlapping bounding boxes will be retained okay. 327 00:21:34,770 --> 00:21:37,860 So let me show you if I set the threshold value high. 328 00:21:49,100 --> 00:21:49,400 And. 329 00:21:54,090 --> 00:21:55,890 Like if I just show you over here. 330 00:22:02,030 --> 00:22:05,330 So now you can see that there are so many overlapping bounding boxes. 331 00:22:05,330 --> 00:22:08,600 Like for dog, they are two for the person they are. 332 00:22:08,600 --> 00:22:11,990 They are multiple hand-bags backpacks are being rejected. 333 00:22:12,020 --> 00:22:14,750 Okay, so there are so many overlapping bounding boxes. 334 00:22:14,750 --> 00:22:20,900 So with the help of this, I would, uh, threshold value for norm expression. 335 00:22:20,930 --> 00:22:25,100 You can vary the, uh, you can vary your results. 336 00:22:25,100 --> 00:22:25,400 You can. 337 00:22:25,400 --> 00:22:26,990 This will affect your results very much. 338 00:22:26,990 --> 00:22:31,670 Like you can see now, uh, if I set the high value of threshold for norm expression. 339 00:22:31,670 --> 00:22:36,230 2%, six Hand-bags six backpacks, two handbags are being detected. 340 00:22:36,230 --> 00:22:36,620 Okay. 341 00:22:37,010 --> 00:22:41,900 So in the same way, you can do object detection on your other images as well. 342 00:22:42,290 --> 00:22:44,690 So you can try this out over here. 343 00:22:47,950 --> 00:22:54,250 See, this is another concept which I want to explain about, uh, norm expression and transport. 344 00:22:54,340 --> 00:22:57,220 So you can try it out with your own images as well. 345 00:22:57,250 --> 00:23:01,540 So now let's test do object detection on the second image over here as well. 346 00:23:03,530 --> 00:23:06,890 So I'm just using the default person's treasure that I treasure now. 347 00:23:07,460 --> 00:23:08,330 Nothing different. 348 00:23:18,200 --> 00:23:20,090 Okay, so we have the results now. 349 00:23:20,090 --> 00:23:20,780 Over here. 350 00:23:22,230 --> 00:23:24,450 So now you can see that we have detected person. 351 00:23:24,450 --> 00:23:25,590 Person. 352 00:23:25,770 --> 00:23:26,160 Okay. 353 00:23:26,160 --> 00:23:27,960 We have detected these two person as well. 354 00:23:27,960 --> 00:23:30,390 And the sports ball is detected as well. 355 00:23:30,390 --> 00:23:33,180 So the results look quite good over here as well. 356 00:23:33,180 --> 00:23:36,060 Like you can see the person over here and over here is quite blurred. 357 00:23:36,060 --> 00:23:38,490 But our model is able to detect it as well. 358 00:23:38,490 --> 00:23:40,860 In this way you can test on other images as well. 359 00:23:40,860 --> 00:23:44,670 So now I will be testing on a video over here as well. 360 00:23:44,670 --> 00:23:45,420 So. 361 00:23:53,140 --> 00:23:55,690 I will just say radio one dot NPY forward. 362 00:23:56,590 --> 00:23:57,010 Okay. 363 00:23:57,010 --> 00:23:58,180 And here. 364 00:23:58,180 --> 00:24:01,510 So now if I want to show the output. 365 00:24:02,500 --> 00:24:03,040 Bye. 366 00:24:03,580 --> 00:24:05,860 So I will just write you the image. 367 00:24:05,860 --> 00:24:09,580 So this will be showing me the live results over here as well. 368 00:24:29,180 --> 00:24:32,060 So now you can see over here this is a wrong detection. 369 00:24:32,060 --> 00:24:36,050 So now you can see that we are able to detect the cars over here as well. 370 00:24:36,050 --> 00:24:37,220 So I'm using CPU. 371 00:24:37,220 --> 00:24:38,690 So the detection is very slow. 372 00:24:38,690 --> 00:24:41,630 Like you can see that uh the inference is very slow. 373 00:24:41,630 --> 00:24:45,380 But you can see that we are able to detect a car over here as well. 374 00:24:45,740 --> 00:24:47,750 So the results look like. 375 00:24:47,750 --> 00:24:47,990 Good. 376 00:24:47,990 --> 00:24:49,070 So in this way. 377 00:24:49,550 --> 00:24:51,800 So this works. 378 00:24:54,730 --> 00:24:58,300 In the same way you can test out on other video as well. 379 00:25:02,240 --> 00:25:06,740 So if you are running on CPU, you can imagine that the processing will be very slow. 380 00:25:06,740 --> 00:25:13,070 But if we tried on GPU, the the frame rate will be very high and the processing will be very fast on 381 00:25:13,070 --> 00:25:13,520 CPU. 382 00:25:13,550 --> 00:25:18,530 The FPS or the frame rate is very low and the processing is very low as well. 383 00:25:21,250 --> 00:25:23,230 Now you can say I've tested on other video. 384 00:25:23,230 --> 00:25:30,070 We are able to detect a person handbag over here as well, and the detection results look quite impressive. 385 00:25:34,060 --> 00:25:38,020 Okay, so in this way you can test on other images and videos as well. 386 00:25:38,020 --> 00:25:42,280 So let's go on and do detection using live webcam feed. 387 00:25:45,820 --> 00:25:48,160 We do object detection using the light webcam feed. 388 00:25:48,190 --> 00:25:55,030 We will set the source to zero if you are using your internal webcam like I am using my HP laptop internal 389 00:25:55,030 --> 00:25:55,630 webcam. 390 00:25:55,630 --> 00:25:59,770 But if you are using an external webcam, then you can set the source 1 or 2. 391 00:26:00,010 --> 00:26:04,420 Okay, but if you have only one external webcam, then you can set the source value to one. 392 00:26:05,020 --> 00:26:05,470 Okay. 393 00:26:05,470 --> 00:26:07,150 And let's run this up now. 394 00:26:25,540 --> 00:26:26,920 When all you can see over here. 395 00:26:26,920 --> 00:26:33,010 I am able to do object detection on the live webcam feed over here as well, so that I'm just running 396 00:26:33,010 --> 00:26:34,240 it on my CPU machine. 397 00:26:34,240 --> 00:26:37,540 So you can see, uh, the detection is very slow. 398 00:26:37,810 --> 00:26:43,330 So that's one of the reason what you can do this object detection on the live webcam feed as well, 399 00:26:43,570 --> 00:26:45,820 like I am doing over here. 400 00:26:45,820 --> 00:26:50,380 And you can see the frame size is already by 640 okay. 401 00:26:53,820 --> 00:26:58,380 So it's 640 is the bid and the 40 is the frame right. 402 00:26:58,410 --> 00:27:01,050 You can see the 640 is the width and frame height. 403 00:27:01,260 --> 00:27:05,310 So let's see I have this uh uh pic of the scissor. 404 00:27:05,310 --> 00:27:08,160 So let's see if it is able to detect the scissor or not. 405 00:27:08,400 --> 00:27:08,640 Okay. 406 00:27:08,640 --> 00:27:09,750 So it's the night. 407 00:27:09,780 --> 00:27:10,950 This is not a knife. 408 00:27:10,950 --> 00:27:11,880 This is a scissor. 409 00:27:12,300 --> 00:27:14,130 And let's see. 410 00:27:14,280 --> 00:27:14,670 Okay. 411 00:27:14,670 --> 00:27:17,490 So now you can see that as decorated as scissor now. 412 00:27:18,000 --> 00:27:18,420 Okay. 413 00:27:18,420 --> 00:27:23,520 So now you can see that here we have overlapping not a cell phone okay. 414 00:27:26,920 --> 00:27:29,200 So now you can see it's going to detect it as Caesar. 415 00:27:29,200 --> 00:27:31,360 And that is fine. 416 00:27:31,360 --> 00:27:34,270 Like now you can see that it's being detected as Caesar. 417 00:27:34,510 --> 00:27:38,110 So in this way you can do object detection on the live webcam feed as well. 418 00:27:38,110 --> 00:27:39,700 And that's all from this tutorial. 419 00:27:39,730 --> 00:27:40,990 Thank you for watching.