1
00:00:00,000 --> 00:00:03,180
One of the most fun and exciting applications of

2
00:00:03,180 --> 00:00:06,600
ConvNet recently has been Neural Style Transfer.

3
00:00:06,600 --> 00:00:12,560
You get to implement this yourself and generate your own artwork in the problem exercise.

4
00:00:12,560 --> 00:00:14,725
But what is Neural Style Transfer?

5
00:00:14,725 --> 00:00:16,710
Let me show you a few examples.

6
00:00:16,710 --> 00:00:18,310
Let's say you take this image,

7
00:00:18,310 --> 00:00:22,140
this is actually taken from the Stanford University not far from

8
00:00:22,140 --> 00:00:25,360
my Stanford office and you want

9
00:00:25,360 --> 00:00:29,940
this picture recreated in the style of this image on the right.

10
00:00:29,940 --> 00:00:33,435
This is actually Van Gogh's, Starry Night painting.

11
00:00:33,435 --> 00:00:36,930
What Neural Style Transfer allows you to do is generated

12
00:00:36,930 --> 00:00:40,620
new image like the one below which is a picture of

13
00:00:40,620 --> 00:00:43,890
the Stanford University Campus that

14
00:00:43,890 --> 00:00:48,580
painted but drawn in the style of the image on the right.

15
00:00:48,580 --> 00:00:53,145
In order to describe how you can implement this yourself,

16
00:00:53,145 --> 00:00:56,644
I'm going to use C to denote the content image,

17
00:00:56,644 --> 00:00:59,545
S to denote the style image,

18
00:00:59,545 --> 00:01:02,990
and G to denote the image you will generate.

19
00:01:02,990 --> 00:01:07,890
Here's another example, let's say you have this content image so let's see

20
00:01:07,890 --> 00:01:13,800
this is of the Golden Gate Bridge in San Francisco and you have this style image,

21
00:01:13,800 --> 00:01:16,740
this is actually Pablo Picasso image.

22
00:01:16,740 --> 00:01:21,120
You can then combine these to generate this image G which

23
00:01:21,120 --> 00:01:26,720
is the Golden Gate painted in the style of that Picasso shown on the right.

24
00:01:26,720 --> 00:01:31,140
The examples shown on this slide were generated by Justin Johnson.

25
00:01:31,140 --> 00:01:37,520
What you'll learn in the next few videos is how you can generate these images yourself.

26
00:01:37,520 --> 00:01:40,050
In order to implement Neural Style Transfer,

27
00:01:40,050 --> 00:01:44,970
you need to look at the features extracted by ConvNet at various layers,

28
00:01:44,970 --> 00:01:47,875
the shallow and the deeper layers of a ConvNet.

29
00:01:47,875 --> 00:01:51,865
Before diving into how you can implement a Neural Style Transfer,

30
00:01:51,865 --> 00:01:54,485
what I want to do in the next video is try to give you

31
00:01:54,485 --> 00:01:59,905
better intuition about whether all these layers of a ConvNet really computing.

32
00:01:59,905 --> 00:02:02,210
Let's take a look at that in the next video.