1
00:00:01,850 --> 00:00:09,560
‫Next is the concept of stride, the first neuron in the convutional layer was looking at these sets

2
00:00:09,560 --> 00:00:10,640
‫of pixels.

3
00:00:12,330 --> 00:00:17,850
‫Now, the next neuron in convutional layer will have a slightly shifted, receptive field.

4
00:00:18,320 --> 00:00:19,010
‫Out the window.

5
00:00:20,460 --> 00:00:25,350
‫This shift, in the view from one neuron to another is called the stride.

6
00:00:27,480 --> 00:00:37,260
‫So if the next neuron is looking at these sets of pixels, you can see that this window, which is for

7
00:00:37,260 --> 00:00:41,190
‫1 the next neuron, is shifted two pixels to the right.

8
00:00:43,050 --> 00:00:50,840
‫So in this scenario, we say that this window has a stride of two each neuron on the convolutional

9
00:00:50,840 --> 00:00:56,160
‫layer will have a window shifted to pixels to the right.

10
00:00:56,880 --> 00:01:05,280
‫So if we continue this, the third neuron will be looking at a set of 25 pixels starting from

11
00:01:05,280 --> 00:01:05,610
‫here.

12
00:01:07,770 --> 00:01:12,840
‫So this will be the field of view for the 3rd neuron and so on.

13
00:01:14,140 --> 00:01:21,460
‫Similarly, a stride of four means that you are taking a jump of four pixels when you look at the next

14
00:01:21,460 --> 00:01:25,020
‫neuron so the second neuron will straight away.

15
00:01:25,120 --> 00:01:30,850
‫Look at these sets of pixels, which is four pixels shifted towards the right.

16
00:01:33,990 --> 00:01:40,650
‫Now, if you use a small stride, then there will be a lot of overlap between these two deceptive

17
00:01:40,650 --> 00:01:41,090
‫Feeds.

18
00:01:42,250 --> 00:01:49,750
‫So if you look at this stride of two, you can see that this three by five rectangle is common between

19
00:01:49,750 --> 00:01:54,790
‫the receptive field of neuron one and deceptive.

20
00:01:54,790 --> 00:01:55,840
‫Firld of neuron 2

21
00:01:56,800 --> 00:02:01,890
‫So we have fifteen pixels common for neuron one and neuron 2

22
00:02:02,640 --> 00:02:04,320
‫But if you have a stride of 4.

23
00:02:05,460 --> 00:02:11,220
‫In that scenario, only this single line is common for neuron one, a neuron to sow only five pixels

24
00:02:11,220 --> 00:02:11,730
‫are common.

25
00:02:12,360 --> 00:02:14,460
‫So there is small overlap.

26
00:02:14,790 --> 00:02:18,510
‫If this trade is large and there is larger overlap, if straight is small.

27
00:02:20,540 --> 00:02:22,500
‫Also if the stride is large.

28
00:02:23,820 --> 00:02:29,940
‫Since the overlap will be less, therefore fewer neurons will be required in the upper layer.

29
00:02:31,640 --> 00:02:39,800
‫So if we have a stride of 4 then the neurons in the upper layer, we'll be less than if we have a stride

30
00:02:39,800 --> 00:02:40,250
‫of two.

31
00:02:42,380 --> 00:02:50,260
‫So the stride will determine the size of upper layer and the amount of overlap and the deceptive fields.