1 00:00:00,210 --> 00:00:00,640 I guys. 2 00:00:00,810 --> 00:00:06,990 Come back in this less than we'll be training a hardhat detector that can be very useful for construction 3 00:00:06,990 --> 00:00:11,010 sites or sites where you need to wear specific PPE gear. 4 00:00:11,550 --> 00:00:12,540 So let's take a look. 5 00:00:12,590 --> 00:00:15,540 So let's open the book 50 and we'll begin to listen. 6 00:00:15,780 --> 00:00:18,000 So this is what we're trying to identify here. 7 00:00:18,420 --> 00:00:20,730 What do these hard hats are being worn? 8 00:00:20,760 --> 00:00:23,820 So you have to identify the hard hats, essentially. 9 00:00:24,660 --> 00:00:27,960 So again, this is a rubber floor notebook that I've adopted here. 10 00:00:27,960 --> 00:00:33,180 So I've heard you run most of these cells already except the treat, except for the infrared cells, 11 00:00:33,180 --> 00:00:35,310 because the model is being trained right now. 12 00:00:36,180 --> 00:00:37,740 However, you can see how long it takes. 13 00:00:37,740 --> 00:00:42,910 The first one takes about 10 seconds and about a minute, and then about 10 seconds here. 14 00:00:42,930 --> 00:00:45,780 These are all set up and install commands. 15 00:00:45,780 --> 00:00:47,840 You don't really have to know what's going on too much. 16 00:00:47,850 --> 00:00:51,780 I mean, it would be good if you wanted to dig deeper and understand exactly what's happening, but 17 00:00:52,200 --> 00:00:55,050 all of these models have the environments. 18 00:00:55,050 --> 00:01:00,660 Settings did different requirements, different parameters you need to set initially, and it can get 19 00:01:00,660 --> 00:01:01,550 quite confusing. 20 00:01:01,560 --> 00:01:09,420 So I suggest if you wanted to understand one of them quite well, maybe take a look at your little vision 21 00:01:09,420 --> 00:01:16,380 five with PI torch, which is coming up quite soon, maybe next to lessons or record that video and 22 00:01:16,380 --> 00:01:22,410 maybe take a look at eficiente detect because efficient detect as well as detection are quite useful 23 00:01:22,410 --> 00:01:22,980 as well. 24 00:01:23,940 --> 00:01:26,140 The others may be less popular. 25 00:01:26,160 --> 00:01:32,190 Our CNN's SSD stores on those popular anymore, but you can still bring those here as well, so it's 26 00:01:32,190 --> 00:01:34,080 up to you whatever you want to use. 27 00:01:34,650 --> 00:01:39,450 I would recommend you a little fish and five for pretty much all of your object detection tests. 28 00:01:40,260 --> 00:01:41,370 So let's move on. 29 00:01:41,370 --> 00:01:42,710 So we've done with the data here. 30 00:01:42,720 --> 00:01:46,470 It's not a very big dataset and you can see the images there. 31 00:01:47,130 --> 00:01:52,260 Then we just create the directory as we place the data in the right directories and then we can start 32 00:01:52,260 --> 00:01:53,700 treating our efficient detect. 33 00:01:54,090 --> 00:01:56,670 We're going into creating an efficient detect D0 zero model. 34 00:01:56,670 --> 00:02:04,380 I believe we set that up to me, which actually drains quite fast because the D0 back backbone is quite 35 00:02:04,380 --> 00:02:05,610 compact and quite small. 36 00:02:06,690 --> 00:02:12,360 So let's take a look and no training here, and you can see this is the training output that we can 37 00:02:12,360 --> 00:02:12,660 see. 38 00:02:12,660 --> 00:02:15,120 So it's already started first epoch here. 39 00:02:15,750 --> 00:02:21,750 It's going, it's going to take quite long, as you can see to me, a 12 percent and has 100 epochs 40 00:02:21,750 --> 00:02:22,080 to go. 41 00:02:22,740 --> 00:02:28,260 So it's actually not quicker to go back and I'm pretty sure it is using you, but I can just double 42 00:02:28,260 --> 00:02:28,680 check. 43 00:02:28,690 --> 00:02:30,010 Yeah, just using GPU. 44 00:02:31,050 --> 00:02:36,120 So either way, let's take a look at what the inference looks like if one of his images and you can 45 00:02:36,120 --> 00:02:38,100 see yes, soon got the helmet right. 46 00:02:39,420 --> 00:02:43,530 I don't know if this is a helmet actually thinks it is, and it might be a helmet. 47 00:02:43,560 --> 00:02:49,100 Maybe it's from a guy who is not properly in the scene, like you just leaning forward from Stanton 48 00:02:49,200 --> 00:02:50,120 standing right here? 49 00:02:50,130 --> 00:02:50,760 Who knows? 50 00:02:51,360 --> 00:02:53,030 But that's it for this lesson. 51 00:02:53,040 --> 00:02:59,040 This lesson did not have a lot going on, and it is just a nice and simple, effective way to train 52 00:02:59,340 --> 00:03:01,890 a hard hat detector object detector. 53 00:03:02,670 --> 00:03:04,200 So thank you for watching. 54 00:03:04,230 --> 00:03:08,520 And stay tuned for a very cool medical image project. 55 00:03:09,000 --> 00:03:13,860 We're going to take a look at the blood cell effect of using the old model. 56 00:03:13,920 --> 00:03:16,770 You you'll efficiency, but it looks fairly well. 57 00:03:16,800 --> 00:03:18,510 So stay tuned for that lesson. 58 00:03:18,600 --> 00:03:19,080 Thank you.