Original subtitles

1

Hello and welcome to the homework of much of one face recognition.

2

So this homework was about building a computer vision application that can detect smells which can be

3

very useful today as indeed some businesses consist of building some computer vision tools to understand

4

customers reactions when for example they're watching a movie.

5

So it is widely used in the cinema industry in the movies industry.

6

Some companies indeed use some computer vision tools to recognize customers expressions when they're

7

watching a movie to understand what makes them smile what makes them sad.

8

What makes them have strong emotions so that they can understand the customers.

9

And then for example recommend some new movies which will boost their emotions.

10

So today we're going to do exactly that but only with the small part.

11

We're going to make a computer vision tool that will detect when a person is smiling.

12

And now I get to say congratulations to the people who managed to make this smile detector but also

13

I want to congratulate the people who tried to do this homework.

14

Trying is also very important by trying always progress.

15

We always learn some things even if we don't succeed.

16

So what matters is to try hard.

17

So on that note we're going to start the solution of this homework and the solution of this homework

18

will include everything from the research phase.

19

Because indeed when you are a computer vision scientist or even a machineries scientist or an AI engineer

20

Well there is always the research phase to look for some solutions.

21

So I included this research phase in the Statoil and this research phase consists of finding the hard

22

cascade that will help us build this smell detector and then of course we'll proceed to the implementation

23

phase and we'll of course use the previous code to be more efficient and then eventually we will watch

24

the result in phase 3 we will notice that this is not optimal and therefore in Phase Four we will try

25

to improve the code improve the computer vision application to make a much more accurate smell detector.

26

So let's do this let's start with Phase 1.

27

The research phase.

28

So as you can see I'm on Google right now.

29

We're going to do our research online to find the right cascade to detect this mouse.

30

And so what I'm going to do.

31

And actually we'll find a place where you have all the heart cascades in case some of you were wondering

32

about that.

33

So what I'm going to do I'm going to type here open see the heart cascade it's actually the first here

34

open city the cascade.

35

Let's do this and let's see what we get.

36

OK.

37

So we had a couple of links.

38

The first one is some open city conditions on the open sea website.

39

That's not what we're interested in.

40

Then the second one is still in the open with the condition but that's still not what we're interested

41

in because that's not where we'll find the heart cascade.

42

However on the third link we might have something interesting.

43

This third link is actually the good herb repository of open city.

44

So if we click on the link What will we find.

45

Well we will find all the different Ahar cascades that are allowed to detect features.

46

Most of them are phase features not only human face features but also as you can see cat face features.

47

This one for example can detect the frontal face of a cat.

48

So if you have a cat you can play with it.

49

And also it can detect some weird things like as you can see Russian plate number.

50

So why not if you're in Russia.

51

Well feel free to practice to make a tool to detect Russian plate numbers.

52

And what do we see in all these Hargus games.

53

We see the heart cascade smiled at X-amount and that's the one we'll use to build our smell detector.

54

But you're going to see that it's not that easy.

55

We will need to change some parameters in the implementation to make it work better.

56

So that's the first step of this homework finding the Harker's get smothered ex-MIL Congratulations

57

if you reached that step that's already exent.

58

So now how do we extract this Hargus gate if some of you try to make Right-Click and then do something

59

like Save Link as well.

60

This will not work.

61

The best way to extract that Hargus gate is to open it.

62

I just clicked on it that should open right now.

63

There we go.

64

And inside you will find the whole heart cascade ex-MIL for the smell.

65

And so what you can do then is copy all this texture and the Epogen way to do this is to scroll down

66

all the way down and then press the shift key and then click again to select everything.

67

Then you copy and paste it in a text editor and you save your file by giving the following name cascade.

68

And of course now that ex-MIL this will work for sure you will make sure I have the right Hargus get

69

X amount as smell.

70

And so that's exactly what I did.

71

I copied that pasted it in a text editor and saved it under this name.

72

And so now what we're going to do is open an eye on that because again I really want to make sure you

73

don't forget to connect to the virtual platform.

74

So there we go.

75

Let's do that right now so we don't forget virtual platform that I remind contains all the packages

76

pre-install to to all the computer vision models that we will implement in this course and then we're

77

going to launch spider and I'm going to show you what I prepared so I prepared actually a folder a homework

78

folder and module one that contains this horrid cascade ex-MIL for the smile that we just made.

79

And so now let's go to Val Explorer.

80

Let's go to the folder that contains your computer vision.

81

It is Zed.

82

I'm taking this for that though because in module 1 I added a new folder a new subfolder that I called

83

homework and that contains this new hard cascade underscores now that SML that I've created this homework

84

folder will be provided in the next tutorial which will have the form of an article and you will get

85

this homework folder that contains the four files here.

86

The first two files you recognize them are X-amount for the eye and the face because we will still be

87

detecting the eye and the face but then it contains just new Hargus gate x.

88

For the smile and the homework solution that is open right now and inside which we will implement the

89

solution.

90

All right.

91

So if you're ready let's proceed to face to the implementation.

92

All right.

93

So as I said in the beginning of this oil we're going to do that efficiently and to do that efficiently

94

we're going to take our face recognition code that we implemented through all the module one where we

95

detected the face and the eye.

96

And we're going to simply copy paste the whole code.

97

I'm copying it right now and I'm going to close this because we don't we won't need it anymore and I'm

98

going to paste this code inside the homework solution.

99

So good news.

100

Now we simply need to add something that will detect a smile in my face and your faces of course.

101

So according to you what do we need to do now.

102

Well as you can see in the first code section we load the Cascades.

103

So as you might guess we need to load the cascade for the smell and that's exactly what we're going

104

to do.

105

But before we do that let's not forget to go back to the homework folder because that's the working

106

directory folder we need to be in that folder because it contains this Melkus Kate.

107

So what I'm going to do now I'm going to copy this line pasted below and replace.

108

I cascade by a new object that will represent the small cascade itself.

109

And so I'm going to rename it.

110

Smile cascade.

111

And of course what do I need to do.

112

I need to change the input here by the right input and the right input is no longer Hargus good.

113

I SML it is hard cascade smal SML.

114

Perfect.

115

Now we have this Marcus Gade.

116

We not only have the SML we also have to smoggiest get great.

117

The next step let's move on to the second code section here where we define that function that will

118

do the detections.

119

And so inside this function we need to add the code that will detect a smell in the face or multiple

120

smells if there are multiple faces and draw the rectangles around the smell.

121

So there we go.

122

Let's do this according to you do we need to do that in the reverential of the face like we did with

123

the eyes.

124

Well the answer is of course yes because we cannot have a smell if we don't have a face.

125

So I'm going to copy the three lines here.

126

That corresponds to the detection of the eyes.

127

And I'm going to paste that here right.

128

So these are the same that correspond to the direction of the eyes.

129

And then we're going to make the right replacement to make the detection of this mouse.

130

So the first thing that we need to do is replace this object here by smells smells it will contain.

131

I remind the coordinates of the upper left corner of the rectangle that will detect this mouse.

132

If there are several smells in the videos that is if there are several faces and so this object contains

133

the coordinates of the upper left corner of the rectangles detecting this mouse and also the width and

134

the height of each of these rectangles.

135

OK.

136

So smells this is not an object of I cascade.

137

But this time smile cascade.

138

And from Smout cascade we used to detect multi-skilled method that will apply on the gray region of

139

interest which is the region of interest of the face with a skin factor on 1.1 and a minimum number

140

of neighbors of three.

141

So that's fine.

142

Let's keep that for now and we'll see what happens.

143

All right then we start this new loop that will iterate through the different eyes but smiles that are

144

in the video.

145

So I'm replacing ice by Smaltz.

146

All right.

147

Good but then we will replace the names of these variables here.

148

We chose x y z w an H for the eyes.

149

Now since we're dealing with a mouse we're going to replace x by a sex e y y s y either will Y S W and

150

H by s h.

151

That's safer.

152

OK.

153

And then therefore for each Smout detector we are going to draw the rectangle around this mouse.

154

So we do that still in the region of interest.

155

This time the colored region of interest because we want to have the original footage with the colors.

156

Then of course we need to replace the X and e y by x x and x y and same here X plus E.W. is replaced

157

by as 6 plus s w and E.W. PCH we replaced by S-W plus S H.

158

Perfect.

159

And now let's choose a different color for this mouse.

160

So let's see we have 2 5 5 0 0 0 2 5 5 and 0 and so now that shows 0 0 and 2.5 that will give us a different

161

color.

162

Perfect We will have three different colors to detect the face that will be in blue.

163

Then the eyes in green and now the smell will be detected in red.

164

Perfect.

165

All right so that's it that's actually done for phase 2 we implemented the code to detect a smile.

166

So now let's have a look at the results.

167

I'm going to execute this code and let's see if it manages to detect correctly some smells.

168

So I'm going to select everything here and press command control plus enter to execute.

169

There we go.

170

Here I am.

171

Hello again happy to see you again and congratulations by the way for those of you who reached this

172

step with all the red rectangles in your face.

173

Don't worry about this.

174

This is not our final answer.

175

We will improve this.

176

Of course we don't have a good smell detector here so we have to do something to improve this indeed.

177

So the face is correct correctly detected that either correctly detected but we see some red rectangles

178

all around my face.

179

That's not good.

180

So according to you what do we have to do.

181

Well the answer is to choose some different parameters some different parameters in our detect ask methods.

182

So that's exactly what we're going to do.

183

And in the end we should have only one red rectangle around my ear around my mouth and only when I'm

184

smiling of course we're making smell detector.

185

So let's do this.

186

I'm going to press Q to quit.

187

I just quit.

188

And now let's proceed to phase four because phase 3 was about watching the result.

189

Let's proceed to face forward to improve the code.

190

So as I said we have to change the parameters and the detect multitasking method.

191

All right.

192

So let's look at them one by one.

193

The first one is air or gray which is a region where we apply the detection according to you.

194

Do we need to change that.

195

Well not really.

196

We're not going to replace it by Gray because then we will see some red rectangles all around my place.

197

You know if I have any shape of object in my place that look like a man will see a red rectangle around

198

it.

199

That will be that good.

200

And believe it or not that will be even worse than what we just got.

201

So no we keep our eye Crais because we want to make the detection and the referential of the face for

202

the simple reason that a smile is in the face.

203

Then second parameter 1.1 during what this is about.

204

This is about the scaling factor.

205

And in fact we're going to replace it by a new number.

206

We're going to choose a larger scaling factor and we're going to increase it to 1.7.

207

All right.

208

This is not what will help the most improve the solution but still it will help a little but the parameter

209

that will definitely make a difference is the number of neighbors we really want to increase the number

210

of neighbors because in some way that the direction has to be much more thorough.

211

If we have a low number of minimal neighbors well anything that looks approximately like a smell will

212

be detected as a smell and that will be not good.

213

And that's why we obtained many rectangles in the previous detection.

214

So we're going to considerably increase this minimum number of neighbors.

215

Now the right number is obtained with experimentation.

216

That's another thing very common when doing some computer vision or machine learning or you always have

217

to experiment.

218

Well this actually goes with research.

219

But here we have to replace this number.

220

And I'm going to give you the right number now.

221

I mean the a number that works very well.

222

That will get us a new one rectangle detecting when I'm smiling when you're smiling.

223

And this number is 22.

224

We increased the number of minimum neighbors from 3 to 22 and that's all we need to do to improve this

225

computer vision solution.

226

And now let's look at the results.

227

I'm going to execute that again.

228

There we go.

229

Executed.

230

Now let's see.

231

All right.

232

So hello again.

233

Well now as you can see I'm not smelling I'm neutral and nothing happens.

234

What does happen when I smile.

235

There we go.

236

We can see a red rectangle detecting detecting when I'm smelling.

237

I'm sorry for the non natural smell but I'm trying to make this work all right here it works very well.

238

We can we can try to still improve this.

239

So I'm going to press Q And I'm even going to increase the minimum number of neighbors of the eyes and

240

I'm also going to remove my mike because I think it's interfering with the detection.

241

So let's see.

242

I'm going to increase the minimum number of neighbors of the eyes to also 22 and we'll see what happens.

243

So there we go.

244

Now we have for the eye a scaling factor of 1.1 and 22 minimum neighbors.

245

And for this now we have a scaling factor of 1.7 and 22 minimum neighbors.

246

Now I'm going to put my mike a little bit away from me but not the force that you can still hear me.

247

I'm going to say like that again.

248

Execute.

249

And now let's see what happens.

250

And here are the results.

251

OK.

252

So first of all we see that the eye detection works much better.

253

I'm trying not to smile so we can indeed see two red red to green rectangles around the eyes and not

254

any around my nose as we could observe before.

255

And right now I'm having a neutral smell and not smelling.

256

And what does happen when I smell

257

we see the red rectangle.

258

Cool.

259

And now what has happened when I don't smile.

260

Know smile it's not

261

smile no smile now no smile.

262

Ok it seems to be working very well.

263

I work much better than before.

264

I'm glad we proved this and the smile is really working well.

265

Not only we see only one rectangle when I'm smiling but also it only appears when I smell.

266

All right perfect so that's done for the homework.

267

Congratulations if you reached that phase.

268

That is if you manage to make this mounded takes you by changing the parameters to make it work.

269

That was not an easy homework we had several things to do several phases to go through.

270

But if you managed to get this no phase detector detector and smell detector well really congratulations.

271

And again for those of you who didn't obtain this but still try it and spend some time trying to get

272

a good smell detector.

273

Well congratulations to you to be relieved.

274

That also made you progress definitely.

275

So now it's actually the end of Mudgal one face recognition.

276

I was very happy to do this for us Mudgal with you.

277

Now we're going to take things at the next level with object detection.

278

We're going to build a state of the art computer vision more all that beats any other model for object

279

detection.

280

It's as is the model single shot multi-book detection.

281

So I look forward to starting Munjal too with you.

282

And until then enjoy computer vision.