1
00:00:00,720 --> 00:00:02,070
Hi and welcome back.

2
00:00:02,550 --> 00:00:09,840
In this section, we'll take a look at all of the practical use cases of guns so you can see how useful

3
00:00:09,840 --> 00:00:11,220
they are in the real world.

4
00:00:11,340 --> 00:00:12,330
So let's take a look.

5
00:00:12,900 --> 00:00:19,500
So here's a list of different industries and scenarios where guns can be quite applicable.

6
00:00:20,130 --> 00:00:26,190
You can think about the creative industries here with generating art and music and other designs as

7
00:00:26,190 --> 00:00:26,790
well as well.

8
00:00:26,790 --> 00:00:32,250
It's not really an industry, but deep fakes, but replicating facial style and video if you're doing

9
00:00:32,250 --> 00:00:35,490
it for nefarious purposes or whether for fun.

10
00:00:35,670 --> 00:00:36,360
Feel free.

11
00:00:37,380 --> 00:00:44,760
Security industry with privacy preserving data, as well as enhancing poor CCTV feeds, there's medical

12
00:00:44,760 --> 00:00:51,100
applications where you can do data augmentation and drug discovery, photography, things like fear,

13
00:00:51,150 --> 00:00:59,100
SAP and so many of those cool apps that do guns or change create who opt out of regular photos.

14
00:00:59,910 --> 00:01:05,790
Also in camera scene enhancement because a lot of those night mode, night modes and computational photography

15
00:01:05,790 --> 00:01:07,440
take advantage of Gans as well.

16
00:01:08,430 --> 00:01:10,680
That there's video and image effects we can do.

17
00:01:10,950 --> 00:01:15,100
The special effects industry can use guns, video games as well.

18
00:01:15,100 --> 00:01:22,260
In Nvidia's deal says technology has been able to upscale images from low risk to high rise, with very

19
00:01:22,260 --> 00:01:27,360
little performance hit, much less performance than actually rendering at that resolution.

20
00:01:27,930 --> 00:01:32,790
Then there's video compression marketing materials with virtual try ons.

21
00:01:33,330 --> 00:01:39,810
Autonomous vehicles as well can take advantage of the guns and even in space and physics to improve

22
00:01:39,810 --> 00:01:40,290
astronomy.

23
00:01:40,320 --> 00:01:41,670
Astronomical images.

24
00:01:42,150 --> 00:01:43,740
So let's take a look at some of these here.

25
00:01:44,490 --> 00:01:49,950
So this is the creative industry and there's a lot of different guns it can send to meeting that.

26
00:01:50,430 --> 00:01:53,160
We also have art that has been created by guns.

27
00:01:53,640 --> 00:01:56,850
These are some of the images of art that was generated by guns.

28
00:01:57,300 --> 00:02:03,540
This is a generated artwork was sold for four hundred and thirty two thousand for some reason or the

29
00:02:03,540 --> 00:02:06,000
other, because I don't think it's that great, to be honest.

30
00:02:06,450 --> 00:02:07,280
That's the ones.

31
00:02:07,830 --> 00:02:09,010
Either way, it's fine.

32
00:02:09,030 --> 00:02:17,400
This other gun saw 40000 as well, but it generated images, altered his logo generation as well.

33
00:02:17,400 --> 00:02:18,360
That Logan.

34
00:02:18,990 --> 00:02:23,580
So you can see there's a number of ways we can use guns in the creative industry.

35
00:02:24,660 --> 00:02:30,150
Next, take a look at deepfakes, so we'll actually be doing and implementing some of these deepfakes

36
00:02:30,150 --> 00:02:31,770
and little lessons in this course.

37
00:02:32,820 --> 00:02:36,330
It's actually quite fun and quite cool, but be careful.

38
00:02:36,510 --> 00:02:42,240
I should give you a warning that it's not good to create deepfakes for cyberbullying purposes or other

39
00:02:42,240 --> 00:02:46,260
nefarious actions, but fine if you're doing it for fun, can go ahead.

40
00:02:46,800 --> 00:02:51,360
So this one is Nicolas Cage imprinted on this actress here.

41
00:02:51,900 --> 00:02:53,760
You can take a look at this one here.

42
00:02:53,790 --> 00:02:59,040
This is a guy talking here, and we just put this face over his face here.

43
00:02:59,700 --> 00:03:04,560
So this is an illustration of how it's being done, and this is arcane again.

44
00:03:04,980 --> 00:03:11,340
So this was this is Tom Cruise in the arc in that Netflix arcane style using Arkin again, and this

45
00:03:11,340 --> 00:03:16,260
is my me and my wife, and different images that are fed into the orchid as well.

46
00:03:16,650 --> 00:03:17,610
So it's pretty cool.

47
00:03:17,620 --> 00:03:18,870
It's pretty cool to experiment.

48
00:03:18,880 --> 00:03:21,330
And what next?

49
00:03:21,330 --> 00:03:26,310
We can take a look at the security industry and see how we can use it to clean up license plates that

50
00:03:26,310 --> 00:03:28,450
have been slightly blurry.

51
00:03:28,800 --> 00:03:31,530
This one here, this is quite blurry, isn't it?

52
00:03:31,890 --> 00:03:34,980
And you can see it's cleaned up to be too much sharper image here.

53
00:03:35,640 --> 00:03:37,860
There's also things like privacy maintenance.

54
00:03:38,280 --> 00:03:44,970
So instead of sharing real data for different academic or research purposes, you can create a synthetic

55
00:03:44,970 --> 00:03:47,880
data out of it that is indistinguishable from the real thing.

56
00:03:48,780 --> 00:03:50,430
Then there's cryptography as well.

57
00:03:51,060 --> 00:03:54,990
You can learn to protect communications with adversarial neural cryptography.

58
00:03:55,530 --> 00:03:59,880
And again, system CCTV footage enhancement, which have shown below here.

59
00:04:01,080 --> 00:04:04,590
Next, you can take a look at medical image medical applications.

60
00:04:05,040 --> 00:04:09,610
So this drug discovery scans can quickly generate novel biological components.

61
00:04:09,610 --> 00:04:15,210
The test hypothesis is that simultaneously, then there's data augmentations.

62
00:04:15,210 --> 00:04:22,470
We can use it to augment medical brain scans so you can generate new datasets, so to improve the brain

63
00:04:22,470 --> 00:04:24,100
disease classifiers as well.

64
00:04:24,120 --> 00:04:28,440
You can see in this example improved from 85 to 92 percent squared.

65
00:04:28,440 --> 00:04:31,410
Good next to good photography.

66
00:04:31,950 --> 00:04:37,350
Things like face up that basically can morph your face into anything you want to convey to the child

67
00:04:37,570 --> 00:04:43,980
or old person with someone else's fears at makeup and to change your nose.

68
00:04:44,760 --> 00:04:45,900
Tons of different things.

69
00:04:46,590 --> 00:04:48,840
I don't use it that much, but I've seen it.

70
00:04:49,330 --> 00:04:54,330
OK, so you can see this is how the scan is actually changing.

71
00:04:54,330 --> 00:04:59,780
This faces here as well as you can do things like change, hairstyles change.

72
00:05:00,110 --> 00:05:06,290
Fearsome male to female is quite interesting and quite weird, sometimes you can remove rain from an

73
00:05:06,290 --> 00:05:07,440
image scene as well.

74
00:05:07,610 --> 00:05:09,170
You can dream scenes.

75
00:05:09,710 --> 00:05:11,900
It's quite useful in things like football.

76
00:05:11,900 --> 00:05:13,280
We wanted to clear up the image.

77
00:05:14,960 --> 00:05:17,100
Also, this is a very cool application.

78
00:05:17,120 --> 00:05:25,100
This is the invidious deep learning supersampling DLC for Short, which allows us to upskill the attorney

79
00:05:25,100 --> 00:05:32,600
type image game, rendering to 4K image and maintain a very high frame rate as well.

80
00:05:33,110 --> 00:05:40,740
You can see in this comparison here this is a good deal SS one at 4K and this is native 4K as well.

81
00:05:41,150 --> 00:05:43,310
And you can see it's virtually almost a.

82
00:05:44,330 --> 00:05:48,170
In fact, I think the deal is this one on the left looks better, in my opinion.

83
00:05:48,890 --> 00:05:52,340
So it's quite interesting and you can see is running at twice the frame rate.

84
00:05:52,430 --> 00:05:53,630
That's quite an achievement.

85
00:05:55,130 --> 00:06:01,910
Next, you can take a look at video compression, and this one was quite famous in late 2020, when

86
00:06:01,910 --> 00:06:09,290
the zupan covered basically had everyone being using the Zoom, and Google meets for calls and video

87
00:06:09,290 --> 00:06:16,040
was showing a demo of how they could actually could this be treated as true 60p video, which was then

88
00:06:16,040 --> 00:06:19,580
a bit bandwidth of 97 kilobits per frame.

89
00:06:20,390 --> 00:06:24,980
If the air compression algorithms they use were able to convert it to this, look at this.

90
00:06:24,980 --> 00:06:32,510
This is like a hundred thousand x the other improvement and decrease in speed.

91
00:06:32,690 --> 00:06:35,780
It's amazing and you can see a video of it here.

92
00:06:36,950 --> 00:06:41,690
This was the original on the left, again reconstructed in the middle, and you can actually see you

93
00:06:41,690 --> 00:06:45,560
can actually move again her own because it's a treaty reconstruction.

94
00:06:46,430 --> 00:06:47,720
Let's take a look at it again.

95
00:06:50,750 --> 00:06:51,800
Pretty cool, isn't it?

96
00:06:55,070 --> 00:06:57,140
Next, we can take a look at this one.

97
00:06:57,170 --> 00:06:59,420
This one is a very, very cool example.

98
00:06:59,420 --> 00:07:05,390
I love this research paper and I encourage you to check it out and uses cycle guide and picks topics

99
00:07:05,780 --> 00:07:12,170
to basically add special effects to images like this so you can augment spaceship in this desert scene

100
00:07:12,170 --> 00:07:12,440
here.

101
00:07:13,400 --> 00:07:17,780
And this floating capsule as well, and it looks amazing.

102
00:07:17,780 --> 00:07:20,660
Looks like something from Hollywood Studio.

103
00:07:21,260 --> 00:07:22,160
Pretty cool, isn't it?

104
00:07:22,260 --> 00:07:22,880
When you can do it.

105
00:07:25,220 --> 00:07:30,080
And then again, you can look at marketing materials so you can see how you can do different clothes,

106
00:07:30,080 --> 00:07:32,690
fittings and images, as well as quite useful.

107
00:07:33,890 --> 00:07:37,130
Then you can look at autonomous vehicles because guns can use.

108
00:07:37,160 --> 00:07:44,750
You can use a lot of things for driving around and guns like simulating environments, inferring included

109
00:07:44,750 --> 00:07:45,290
objects.

110
00:07:45,300 --> 00:07:52,250
If a pedestrian is being partially blocked, you can use guns to enhance that in betting as well for

111
00:07:52,250 --> 00:07:57,080
missing areas or noise and sensors and super resolution again as well.

112
00:07:57,110 --> 00:07:59,150
So you can see how useful guns are.

113
00:07:59,630 --> 00:08:04,880
This is an example of investing here, so imagine you were driving and something was blocking your view

114
00:08:04,880 --> 00:08:05,510
like this.

115
00:08:06,350 --> 00:08:13,100
You can use it against infer what's behind it, because that's what us as humans do a lot we do know,

116
00:08:13,100 --> 00:08:20,360
like from a prior knowledge, like if a truck is blocking faster road, we can know that we can instinctively

117
00:08:20,360 --> 00:08:25,220
know that there's different things that's going to be behind there based on what we've seen before.

118
00:08:25,520 --> 00:08:27,770
So guns to a similar thing as well.

119
00:08:29,030 --> 00:08:33,140
And lastly, in space and physics, you can use guns as a road.

120
00:08:33,440 --> 00:08:40,730
And Cosmo Gun was used to study dark matter as well as guns can improve astronomical images, simulate

121
00:08:40,730 --> 00:08:43,070
gravitational lensing for dark matter research.

122
00:08:43,070 --> 00:08:44,330
So it's all quite cool.

123
00:08:45,020 --> 00:08:48,470
So that's it for this lesson on guns.

124
00:08:48,950 --> 00:08:55,490
What we're going to do now, we're going to go back into Google CoLab and use PyTorch and Keros to do

125
00:08:55,490 --> 00:08:57,530
a number of different Gantt experiments.

126
00:08:57,560 --> 00:09:04,200
What we're going to take a look at is generating, firstly, the amnesty dataset from guns to basic.

127
00:09:04,220 --> 00:09:05,570
That's the gun one and one.

128
00:09:06,260 --> 00:09:13,790
Then we're going to use super resolution on Keros to do an image enhancement.

129
00:09:13,790 --> 00:09:17,370
With that, we're going to generate animate characters with style.

130
00:09:17,370 --> 00:09:24,260
Again, we're going to convert horses to zebras using cycle gun, and then we're going to use kid again

131
00:09:24,620 --> 00:09:27,420
to convert an image to the akin style.

132
00:09:27,470 --> 00:09:30,770
The Netflix series are going to have a very unique artistic style.

133
00:09:31,250 --> 00:09:34,250
We're going to convert any image you want to that style.

134
00:09:34,670 --> 00:09:36,840
So I'll see you in the next few lessons.

135
00:09:37,250 --> 00:09:40,640
We're going to be quite fun and exciting, so I hope you enjoyed this.

136
00:09:40,760 --> 00:09:41,180
Thank you.