1 00:00:00,990 --> 00:00:09,000 Now in this section we are going to discuss the architecture of some popular CNN models which one image 2 00:00:09,000 --> 00:00:12,980 classification competitions in the past. 3 00:00:12,990 --> 00:00:15,810 There are two reasons for doing this. 4 00:00:15,810 --> 00:00:23,160 First is we want to understand these architectures because this will build our intuition as to what 5 00:00:23,430 --> 00:00:27,080 is a good CNN model architecture. 6 00:00:27,390 --> 00:00:35,400 And the second reason is that these green models can be used by us in our software without retraining 7 00:00:36,720 --> 00:00:44,460 these models with the architecture and the trained weight can be downloaded as part of the Get Us library 8 00:00:44,460 --> 00:00:44,790 only. 9 00:00:46,380 --> 00:00:51,330 But how will these train models which are trained on other data set. 10 00:00:51,660 --> 00:00:52,870 How can these be used. 11 00:00:52,870 --> 00:00:55,590 What our classification problem. 12 00:00:55,590 --> 00:00:59,660 We will see the answer to this question in the coming lectures. 13 00:00:59,910 --> 00:01:08,400 Before we discuss these architectures I briefly tell you about the image net competition which has given 14 00:01:08,400 --> 00:01:11,330 us these popular architectures. 15 00:01:11,580 --> 00:01:19,500 So image that competition which is known as Eilis we are see for short stand for image net. 16 00:01:19,590 --> 00:01:23,910 Large scale visual recognition challenge. 17 00:01:23,910 --> 00:01:29,610 This challenge was held between 2010 and 2017 every year. 18 00:01:29,610 --> 00:01:36,810 Participants in this challenge were given a data set of images and they had to classify those images 19 00:01:36,930 --> 00:01:41,230 into several labeled in some of the competitions. 20 00:01:41,240 --> 00:01:49,100 The dataset had over a million observations and the classes to be identified were intelligent. 21 00:01:49,460 --> 00:01:53,870 So it was a very large scale competition. 22 00:01:53,900 --> 00:02:01,880 The goal of this challenge was to promote the mind of better computer vision techniques and also to 23 00:02:01,880 --> 00:02:03,880 benchmark the state of the art. 24 00:02:04,580 --> 00:02:14,030 So the winner of each year that conventional network was the benchmark network it was considered as 25 00:02:14,030 --> 00:02:15,440 the best of it time 26 00:02:18,910 --> 00:02:21,870 so here are some popular CNN architectures. 27 00:02:21,940 --> 00:02:27,060 Some of these also won the idealist VRC challenge. 28 00:02:27,730 --> 00:02:31,050 The first one here is Leonard. 29 00:02:31,270 --> 00:02:37,150 This was the oldest and most popular CNN architecture. 30 00:02:37,150 --> 00:02:43,890 It has only 60000 parameters and if you look at it it was made in nineteen ninety eight. 31 00:02:43,930 --> 00:02:52,270 So even in 1998 finalist no neural networks were dead and they were gaining popularity but a major breakthrough 32 00:02:52,270 --> 00:03:01,480 came in 2012 when Alex Nate was able to achieve very high accuracy on previous image classification 33 00:03:01,480 --> 00:03:02,050 problems. 34 00:03:04,380 --> 00:03:13,200 This raised the interest of people in CNN and in 2013 the chef Nate won the Eilis. 35 00:03:13,230 --> 00:03:18,130 We asked challenge and it was a conversational network. 36 00:03:18,900 --> 00:03:26,670 It was able to achieve accuracy rate of nearly 85 percent in 2014. 37 00:03:26,670 --> 00:03:30,090 We got two very popular architectures. 38 00:03:30,180 --> 00:03:37,080 One is Google it and the other is we did it google it was the winner. 39 00:03:37,080 --> 00:03:45,750 We did it net was runner up google it had four million parameters as compared to we did in it which 40 00:03:45,750 --> 00:03:50,770 had 138 million parameters in 2005. 41 00:03:50,820 --> 00:03:55,140 The winner was resonate and in 2016 and 17. 42 00:03:55,140 --> 00:04:01,260 Also there were two other architectures which won but these are the most popular architectures that 43 00:04:01,260 --> 00:04:03,350 you should know about. 44 00:04:03,510 --> 00:04:06,990 We look at Leonard Google it. 45 00:04:07,260 --> 00:04:09,810 And we did it in more detail.