Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated: 1 00:00:00,500 --> 00:00:06,610 Hello and welcome to the last step of this module to object detection before we start the homework. 2 00:00:06,650 --> 00:00:12,140 But here we go with the last step and this last step is going to be about the training of the SSD. 3 00:00:12,290 --> 00:00:18,260 So we're not going to implement the whole code that we'll try and the SSD simply because it's a huge 4 00:00:18,260 --> 00:00:18,720 code. 5 00:00:18,740 --> 00:00:23,630 And besides you're going to see that you're going to need specific requirements to be able to run that 6 00:00:23,630 --> 00:00:24,320 training. 7 00:00:24,500 --> 00:00:29,630 So this tutorial is going to be about explaining how you could train the SSD. 8 00:00:29,630 --> 00:00:32,160 So I'm going to show you the file how to execute it. 9 00:00:32,300 --> 00:00:36,700 And mostly I'm going to show you the data set on which the SSD was trained. 10 00:00:36,890 --> 00:00:42,920 So in case you want to train the SSD with some other object you would understand the approach of how 11 00:00:42,920 --> 00:00:43,570 to do it. 12 00:00:43,730 --> 00:00:48,530 But remember you're going to see that you're going to need some pretty advanced system. 13 00:00:48,650 --> 00:00:55,370 So as you can see right now I'm on a web page and this is the web page that contains the data sets on 14 00:00:55,370 --> 00:00:57,080 which the SS was trained. 15 00:00:57,080 --> 00:00:59,320 This dataset is not image net. 16 00:00:59,330 --> 00:01:04,520 If you were hoping for that but the Pascal visual object classes dataset. 17 00:01:04,520 --> 00:01:11,030 So if you remember in the code we saw Virk the OSI you know for classes to have the mapping between 18 00:01:11,030 --> 00:01:12,840 the classes and integers. 19 00:01:12,980 --> 00:01:20,230 Well Varg stands for visual object classes and this is already a huge dataset not as huge as image net 20 00:01:20,300 --> 00:01:24,380 but a huge one that you will see contains lots and lots of images. 21 00:01:24,650 --> 00:01:29,840 So you can find this page at this address but don't worry the next it will be an article which you will 22 00:01:29,840 --> 00:01:37,580 find the homework folder that will contain all the tools to train the SSD Plus the address of this website 23 00:01:37,790 --> 00:01:40,730 so that you can have a look at this better. 24 00:01:40,730 --> 00:01:47,600 All right so let's go through this web page so as you can see the Pascals Virk project provides standardized 25 00:01:47,690 --> 00:01:49,930 image data sets for object recognition. 26 00:01:49,940 --> 00:01:56,010 So exactly what we're doing and besides provides a common set of tools for accessing the data set and 27 00:01:56,010 --> 00:01:57,010 annotations. 28 00:01:57,050 --> 00:02:03,230 So it's not only a data set it's also an advanced structure of data set that helps a lot for the training. 29 00:02:03,350 --> 00:02:07,370 And also you can see that there were some challenges but now finished. 30 00:02:07,430 --> 00:02:11,460 But anyway it is definitely useful data set to train the SSD. 31 00:02:11,460 --> 00:02:18,180 And so now if we have a closer look at what these days that are well we need to scroll down to that. 32 00:02:18,240 --> 00:02:25,610 What challenges from 2005 to 2012 and here are the data sets you can see there's several of them from 33 00:02:25,670 --> 00:02:33,440 2005 to 2012 and the ones we have the ones on which the SSD was trans and I'm going to show you the 34 00:02:33,440 --> 00:02:39,150 code and the data sets themselves and how you can execute the code to train the SSD on these two data 35 00:02:39,150 --> 00:02:39,780 set. 36 00:02:39,920 --> 00:02:45,670 These two data sets are the Vark 2007 and the Vok 2012. 37 00:02:45,980 --> 00:02:51,310 So let's have a closer look at where they are right now and we need to scroll down even more. 38 00:02:51,440 --> 00:02:54,390 Actually at the bottom of the page there we go. 39 00:02:54,650 --> 00:02:58,640 So here you have this table that's the most important part of this page. 40 00:02:58,640 --> 00:03:04,350 That's where you can see how the data sets are structured and made and what they contain exactly. 41 00:03:04,700 --> 00:03:05,990 So let's have a look at the first one. 42 00:03:05,990 --> 00:03:14,060 Remember we have the 2007 data set and 2012 sort of the 2007 data set contains 20 classes. 43 00:03:14,060 --> 00:03:20,560 Remember I told you that the pre-trained the model we have can detect between 30 to 40 object. 44 00:03:20,630 --> 00:03:27,530 Well among these objects we have some persons Well the S's can detect any person in a video or in some 45 00:03:27,530 --> 00:03:36,080 images then several animals some birds cats cows dogs horses sheeps some vehicles airplanes bicycles 46 00:03:36,080 --> 00:03:41,750 boats buses cars you can do some computer vision for some driving car if you have a self-driving car 47 00:03:41,750 --> 00:03:48,530 business I would highly recommend to integrate as is the model for the computer vision inside yourself 48 00:03:48,650 --> 00:03:52,450 in cars are also motorbikes and trains of course. 49 00:03:52,670 --> 00:04:00,770 And then we have some other objects like some bottles chairs dining table potted plants sofas and TV 50 00:04:00,890 --> 00:04:01,920 monitors. 51 00:04:01,940 --> 00:04:02,300 All right. 52 00:04:02,300 --> 00:04:04,850 And I think we have even more but there you go. 53 00:04:04,850 --> 00:04:07,220 You can see you have plenty of objects to detect. 54 00:04:07,220 --> 00:04:09,380 So really really useful. 55 00:04:09,380 --> 00:04:19,160 And now if we go to the 2012 dataset of 2012 we can see we also have 20 classes and the data contains 56 00:04:19,400 --> 00:04:27,560 eleven thousand five hundred and thirty images containing 27 450 animated objects so we're going to 57 00:04:27,560 --> 00:04:33,640 have a look at these objects specifically because as I told you I've prepared for you a folder to training 58 00:04:33,640 --> 00:04:37,990 as is the folder that contains not only the code but also these two data sets. 59 00:04:38,000 --> 00:04:40,470 We're going to have a clear look at images. 60 00:04:40,490 --> 00:04:42,100 So we're going to look at some of them. 61 00:04:42,200 --> 00:04:47,540 And so what I'm going to do right now is walk you through this folder that I've prepared. 62 00:04:47,540 --> 00:04:50,190 So let's open Anaconda up. 63 00:04:50,240 --> 00:04:50,810 There we go. 64 00:04:50,810 --> 00:04:56,660 We need to connect the virtual platform of course because the training is made with by torch again and 65 00:04:56,660 --> 00:05:00,840 some other tools that are preinstalled individual them. 66 00:05:00,850 --> 00:05:01,750 So there we go. 67 00:05:01,810 --> 00:05:06,670 Let's not forget to connect to the virtual platform applications on virtual platform. 68 00:05:06,670 --> 00:05:07,520 There we go. 69 00:05:07,600 --> 00:05:13,960 And now we launch spider so spiders launching and I'm going to show you this training as is the four 70 00:05:13,960 --> 00:05:15,400 that I've prepared for you. 71 00:05:15,400 --> 00:05:22,870 I put it in the module to object detection color which I'm going to go to right now so I'll explore 72 00:05:23,110 --> 00:05:28,570 it on my desktop that's the computer vision it is that folder and if we go to module 2 you'll find a 73 00:05:28,570 --> 00:05:29,590 new folder. 74 00:05:29,720 --> 00:05:35,590 This folder training as is d that contains all the tools and data sets for you to train the SSD. 75 00:05:35,830 --> 00:05:38,760 But I have to specify something very important now. 76 00:05:38,920 --> 00:05:45,290 Remember at the beginning of this oil I told you that you need an advanced system to train the SSD while 77 00:05:45,430 --> 00:05:49,140 this advance system is simply to have kuda on your machine. 78 00:05:49,180 --> 00:05:55,900 Kuda is like an accelerator that is connected to every geographic cards and that allows to speed up 79 00:05:55,960 --> 00:05:58,990 considerably the training computations. 80 00:05:58,990 --> 00:06:04,880 Therefore the training implementations that were made are only compatible with kuda. 81 00:06:04,890 --> 00:06:10,870 I'm going to tell you a fun fact now if there was an implementation of the training implemented without 82 00:06:10,870 --> 00:06:16,240 kuda I can tell you for sure the training would take several months or even years. 83 00:06:16,240 --> 00:06:17,440 This is not a joke. 84 00:06:17,500 --> 00:06:22,370 The training of the SSD is first of all made with thousands and thousands of images. 85 00:06:22,540 --> 00:06:27,390 But if you train that without kuda it would definitely take several months. 86 00:06:27,400 --> 00:06:33,460 You can not trained the SSD on your CPQ you have to train it with a GP you and the best you can find 87 00:06:33,520 --> 00:06:34,390 is kuda. 88 00:06:34,480 --> 00:06:40,990 So just to set things clear this implementation is only kuda compatible and therefore you will only 89 00:06:40,990 --> 00:06:47,320 be able to execute the code if you have an NDA Geographic card with kuda and labels on your machine. 90 00:06:47,320 --> 00:06:48,720 But even if that's the case. 91 00:06:48,820 --> 00:06:53,550 While it would take some time and you would not have your SSD trained in the next hour. 92 00:06:53,860 --> 00:06:58,960 But I think it's still interesting for you to see how you can run the training and therefore that's 93 00:06:58,960 --> 00:07:00,650 exactly what I'm going to show you right now. 94 00:07:00,790 --> 00:07:08,410 But first let me walk you through this training SSD folder starting with the data so you can see we 95 00:07:08,410 --> 00:07:11,170 have this data folder that contains several files. 96 00:07:11,200 --> 00:07:17,650 Here you can have the scripts which are shellscript and which allow to download the datasets Vogue 2007 97 00:07:17,670 --> 00:07:19,180 and Vok 2012. 98 00:07:19,360 --> 00:07:27,940 So to do that you simply need to open a terminal and then type S H space Virk 2007 dot S H then execute 99 00:07:28,270 --> 00:07:35,350 and this will download word 2007 and then you can type another command as a space walk 2012 that s h 100 00:07:35,740 --> 00:07:37,880 and then press enter to the queue to download. 101 00:07:37,910 --> 00:07:41,790 Also 2012 but I've already done that for you. 102 00:07:41,800 --> 00:07:49,180 I've downloaded this dataset and here they are there in this folder Vark dev kit and you have Virk 2007 103 00:07:49,430 --> 00:07:51,080 and Vark 2012. 104 00:07:51,460 --> 00:07:53,310 So let's have a look at Vocht 2007. 105 00:07:53,320 --> 00:07:58,390 First you have several subfolder but the one I want to show you is this one. 106 00:07:58,480 --> 00:08:02,620 GPL images which contains lots and lots of images. 107 00:08:02,830 --> 00:08:05,630 As you can see let's open randomly. 108 00:08:05,650 --> 00:08:09,070 Some of them so let's see this one for example number 8. 109 00:08:09,070 --> 00:08:09,980 It's a chair. 110 00:08:10,020 --> 00:08:13,220 All right so that's the ground truth for the chair. 111 00:08:13,240 --> 00:08:19,510 Remember Kyrios intuition lectures need to start with the ground truth to ground truth is the truth 112 00:08:19,600 --> 00:08:23,370 that there is in fact a chair right here on this image. 113 00:08:23,460 --> 00:08:28,450 So that's the ground truth that will be compared to the prediction of the SS doing the training just 114 00:08:28,450 --> 00:08:32,590 before a possible error is back propagated into the neural network. 115 00:08:32,590 --> 00:08:35,350 All right so let's close this let's open another one. 116 00:08:35,350 --> 00:08:36,380 Number 14. 117 00:08:36,430 --> 00:08:41,470 It's a car a yellow cab probably in New York I would say maybe. 118 00:08:41,560 --> 00:08:43,340 Or some other cities in the US. 119 00:08:43,450 --> 00:08:44,210 But there we go. 120 00:08:44,230 --> 00:08:49,720 That's to detect a car so it's a ground truth for a car number 31 then a train. 121 00:08:49,720 --> 00:08:50,170 There we go. 122 00:08:50,170 --> 00:08:52,820 You can even detect train. 123 00:08:52,940 --> 00:08:55,030 And let's open a few more. 124 00:08:55,030 --> 00:08:58,030 So that's a pigeon to detect birds. 125 00:08:58,060 --> 00:08:59,620 Ground Truth for the bird. 126 00:08:59,920 --> 00:09:03,260 And now 53 a cat cute little cat. 127 00:09:03,520 --> 00:09:05,920 And one last one hundred. 128 00:09:05,950 --> 00:09:07,730 Well let's go further. 129 00:09:07,900 --> 00:09:11,290 That's open the number one thousand five hundred eighty six. 130 00:09:11,290 --> 00:09:14,000 And that's a beautiful horse jumping. 131 00:09:14,140 --> 00:09:18,240 And speaking of horses well that's going to be the homework for this section. 132 00:09:18,280 --> 00:09:24,070 It's going to be it's going to be a fun homework for you to relax and enjoy playing some object detection 133 00:09:24,130 --> 00:09:26,470 and some really really nice videos. 134 00:09:26,500 --> 00:09:30,690 We're going to do some action on beautiful horses running on a field. 135 00:09:30,760 --> 00:09:31,600 So there we go. 136 00:09:31,600 --> 00:09:32,610 You have the object. 137 00:09:32,630 --> 00:09:38,380 And now let's have a look at the other data set because the training is made on the two data sets at 138 00:09:38,380 --> 00:09:39,320 the same time. 139 00:09:39,580 --> 00:09:42,990 So have to scroll back up. 140 00:09:43,210 --> 00:09:44,520 There we go. 141 00:09:44,620 --> 00:09:48,300 Varg 2017 and now let's have a look at 2012. 142 00:09:48,580 --> 00:09:50,010 Same subfolder is. 143 00:09:50,230 --> 00:09:53,950 But this time with different images. 144 00:09:53,970 --> 00:09:54,250 All right. 145 00:09:54,250 --> 00:09:55,160 Here they are. 146 00:09:55,330 --> 00:10:07,550 So the 2012 we actually have images of several 2007 2008 and probably 2009 there we go and up to 2012. 147 00:10:07,550 --> 00:10:11,510 So let's open any of them for example this one from 2010. 148 00:10:11,720 --> 00:10:15,020 Beautiful dog with a somewhat suspicious look. 149 00:10:15,260 --> 00:10:16,490 But I love dogs. 150 00:10:16,490 --> 00:10:21,160 I hope you liked the dog bouncing on the field as we detected during this much too. 151 00:10:21,170 --> 00:10:22,340 I really like this Doug. 152 00:10:22,560 --> 00:10:28,390 So let's have let's open another one let's open one image from 2011. 153 00:10:28,520 --> 00:10:29,210 There we go. 154 00:10:29,210 --> 00:10:30,010 This one. 155 00:10:30,230 --> 00:10:34,640 Well two nice persons probably a mother with her daughter. 156 00:10:34,740 --> 00:10:35,700 So very nice. 157 00:10:35,810 --> 00:10:41,160 Let's open one from 2012 now and that will be our last. 158 00:10:41,180 --> 00:10:44,100 So there we go the less ground truth we have. 159 00:10:44,270 --> 00:10:47,000 Oh another kid and we can see several persons. 160 00:10:47,000 --> 00:10:48,160 This one is actually interesting. 161 00:10:48,160 --> 00:10:50,640 There are several ground truth in this image. 162 00:10:50,720 --> 00:10:55,240 So maybe let's open the last one to look for a last object. 163 00:10:55,240 --> 00:11:02,810 So that's another person and another person so probably all the images from 2012 are persons. 164 00:11:02,820 --> 00:11:10,490 Now here we have a horse and a person so that's basically combining several different classes for the 165 00:11:10,760 --> 00:11:15,660 day to be able to recognize several objects in an image. 166 00:11:15,660 --> 00:11:21,300 All right so we're going to stop here and now now that we're done with the dataset I'm going to show 167 00:11:21,300 --> 00:11:24,080 you the code and I'm going to show you how to run the code. 168 00:11:24,780 --> 00:11:32,490 First let me just scroll back up to find my way back to the folder. 169 00:11:32,850 --> 00:11:33,740 There we go. 170 00:11:33,780 --> 00:11:40,370 Scripts done data done and going back to the main training as the folder. 171 00:11:40,530 --> 00:11:41,210 All right. 172 00:11:41,530 --> 00:11:46,930 So in this folder you also have besides the data you also have the layers which was the same folder 173 00:11:46,960 --> 00:11:49,030 as before when we made the detection. 174 00:11:49,030 --> 00:11:54,260 So that's because in order to train the SSD we need some of the functions and modules of the SSD. 175 00:11:54,370 --> 00:11:59,290 Then we have the SSD implementation that contains all the architecture of the SSD Anchorage to have 176 00:11:59,290 --> 00:12:00,100 a look at this. 177 00:12:00,370 --> 00:12:06,320 And we have our trained file which I'm going to open right now because data from this file that we're 178 00:12:06,340 --> 00:12:17,780 going to execute the training of the SSD on these two datasets Word 2007 and book 2012. 179 00:12:17,810 --> 00:12:18,870 All right. 180 00:12:19,200 --> 00:12:25,110 And then we have you Teal's which contains some tools for the training like augmentations so augmentations 181 00:12:25,110 --> 00:12:31,580 are some ways to increase the amount of images so that we can have even more material to train our SD 182 00:12:31,590 --> 00:12:34,730 on even if we have already lots and lots of images. 183 00:12:34,800 --> 00:12:40,020 You can actually check that in our deep learning course we have some tutorials explaining in more details 184 00:12:40,320 --> 00:12:45,900 how Image augmentations work and then finally we have this for that that contains some weights. 185 00:12:45,930 --> 00:12:51,030 So make sure to understand that these are not some weights of some pre-trained model because actually 186 00:12:51,030 --> 00:12:56,600 right now we're doing some training but these are just some initialized weights. 187 00:12:56,700 --> 00:13:02,460 You imagine that they need to be initialized in some way a way that should be compatible with the future 188 00:13:02,460 --> 00:13:04,510 of data the way weights during the training. 189 00:13:04,710 --> 00:13:10,620 So these are the weights and now we're going to go through this train file and mostly we're going to 190 00:13:10,620 --> 00:13:14,110 execute the code to show you how the training is done. 191 00:13:14,160 --> 00:13:18,390 It's actually going to be very easy we simply need to select the whole code and then press command or 192 00:13:18,390 --> 00:13:20,280 control press and to to execute. 193 00:13:20,280 --> 00:13:25,040 But before we do that I would just like to show you some of the arguments here. 194 00:13:25,170 --> 00:13:30,660 So basically these are some parcels containing all the arguments that you can change if you want to 195 00:13:30,660 --> 00:13:32,220 change the way the training will be done. 196 00:13:32,220 --> 00:13:38,160 So for example you can change the argument related to the learning rate here the default learning rate 197 00:13:38,160 --> 00:13:45,750 is 0.01 but you can change it to 0.05 for example if you want to try a different training then you can 198 00:13:45,750 --> 00:13:48,220 also change the parameter for the momentum. 199 00:13:48,330 --> 00:13:50,810 The way to the gamma parameter. 200 00:13:50,940 --> 00:13:56,400 Well these are all the parameters that you can change the hyper parameters to do some type of parameter 201 00:13:56,400 --> 00:13:57,860 tuning of the training. 202 00:13:57,930 --> 00:14:03,600 But again remember that training is taking lots and lots of time so it's mostly for the research and 203 00:14:03,600 --> 00:14:06,990 developers working on the state of the art the models. 204 00:14:06,990 --> 00:14:09,410 That's not for us to do right now anyway. 205 00:14:09,600 --> 00:14:12,920 But then you have some other parameters I mentioned. 206 00:14:12,990 --> 00:14:20,040 Image net which is another big huge dataset on which you can train your DeBerry morals and computer 207 00:14:20,040 --> 00:14:27,710 vision models and therefore if you want to train your SSD on a different data set then Virk 2007 invoked 208 00:14:27,730 --> 00:14:28,870 2012. 209 00:14:29,070 --> 00:14:37,390 Well here is the parameter related to that which contains the path to the dataset data if get that exactly 210 00:14:37,400 --> 00:14:45,250 the path data as this father and then there it is this folder containing Vok 2007 invoke 2012. 211 00:14:45,520 --> 00:14:48,280 All right so let me go back. 212 00:14:48,540 --> 00:14:50,160 So basically that's how it works. 213 00:14:50,160 --> 00:14:55,530 You know you have several hyper parameters you can change them to experiment different trainings and 214 00:14:55,530 --> 00:14:59,750 the rest is some implementation of other functions like. 215 00:14:59,760 --> 00:15:04,590 Exactly and wait and it function that initializers the way it's the proper way. 216 00:15:04,590 --> 00:15:09,990 So we're going to see that in more details in module 3 because in modules we will implement from scratch 217 00:15:10,200 --> 00:15:11,490 the training of the Ganns. 218 00:15:11,510 --> 00:15:17,100 Generally our research on that works because this will be possible to do without kuda and without thousands 219 00:15:17,100 --> 00:15:18,080 of lines of code. 220 00:15:18,240 --> 00:15:21,120 So you have several functions and mostly we have. 221 00:15:21,210 --> 00:15:28,470 There you go the train function that does the training of the SSD on all the images that are contained 222 00:15:28,560 --> 00:15:30,390 in the Vok folders. 223 00:15:30,390 --> 00:15:32,840 So there we go now to finish the tutorial. 224 00:15:32,850 --> 00:15:39,740 I'm going to show you how you can execute this file so very simply you just select the whole code again. 225 00:15:39,840 --> 00:15:45,660 Make sure that you have an empty Geographic card and kuda enabled and that's not the case you will get 226 00:15:45,780 --> 00:15:47,460 an error and be relieved. 227 00:15:47,490 --> 00:15:50,890 Even if we had a non a version of this training. 228 00:15:51,070 --> 00:15:53,150 Well it would take months or years. 229 00:15:53,190 --> 00:15:55,980 And therefore this would be completely useless. 230 00:15:56,190 --> 00:16:02,310 And that's exactly the reason why the developers have not implemented and none could have version of 231 00:16:02,320 --> 00:16:03,440 the training. 232 00:16:03,600 --> 00:16:05,910 That is because it would be completely useless. 233 00:16:06,030 --> 00:16:08,570 So I'm going to you now. 234 00:16:08,580 --> 00:16:09,500 There we go. 235 00:16:09,510 --> 00:16:16,190 The train is on its way and it's basically going to take several hours so we're just going to stop here. 236 00:16:16,200 --> 00:16:21,750 It was just to show you how the training works how the data sets were structured and to show you a little 237 00:16:21,750 --> 00:16:24,390 behind the scene how the SS is trained. 238 00:16:24,570 --> 00:16:31,160 And now we hope you understand why we had to use a pre-trained model to do our detection. 239 00:16:31,170 --> 00:16:33,800 All right so now we're going to move on to the homework. 240 00:16:33,960 --> 00:16:40,170 It's going to be a fun and exciting homework about detecting some beautiful horses running on a field. 241 00:16:40,230 --> 00:16:41,490 I can't wait to show you this. 242 00:16:41,490 --> 00:16:43,270 And until then enjoy computer vision. 25548