1
00:00:02,780 --> 00:00:04,670
Next is Viji 16.

2
00:00:06,020 --> 00:00:10,860
This was The Runner-Up in 2014, a list VRC challenge.

3
00:00:11,720 --> 00:00:17,630
This was a very big architecture and it had 138 million parameters to be trained.

4
00:00:18,440 --> 00:00:20,660
So it took a lot of time to train.

5
00:00:22,010 --> 00:00:28,190
If you look at the architecture instead of convolutional layers, it does convolutional blocks.

6
00:00:29,000 --> 00:00:37,640
And one block consists of several congressional layers and a max pulling layer at the end to this convolutional

7
00:00:37,640 --> 00:00:42,920
block has to convolutional layers and one max pooling layer.

8
00:00:44,330 --> 00:00:48,440
Next convolutional block has two corners, layers and IMAX Willinger.

9
00:00:49,730 --> 00:00:52,370
And then we are going to have a congressional block.

10
00:00:53,720 --> 00:00:59,650
The last two convolutional blocks have three convolutional lives instead of two congressional leaders.

11
00:01:02,120 --> 00:01:11,030
After processing a two 24 hour to 24 pixel image at the end of all of these convolutional layers, we

12
00:01:11,030 --> 00:01:15,760
had 512 feature maps of seven by seven sites.

13
00:01:17,750 --> 00:01:25,820
These feature maps were fed into a fully connected neural network with so many neurons.

14
00:01:27,710 --> 00:01:34,520
At the end of it, there were 1000 output neurons with softmax activation.

15
00:01:35,570 --> 00:01:44,360
This is because Viji sixteen was run on a problem in which there were 1000 classes into which images

16
00:01:44,360 --> 00:01:45,440
were to be classified.

17
00:01:46,070 --> 00:01:47,840
So it was a huge database.

18
00:01:48,130 --> 00:01:55,090
It had over a million images and those images were to be classified and do 1000 classes.

19
00:01:55,670 --> 00:01:57,980
Classes were of dog breeds.

20
00:01:58,010 --> 00:01:59,240
God breeds a tetra.
