Convolutional Neural

Networks_Ex
Dr. Dinesh Kumar Vishwakarma
Professor, Department of Information Technology
Delhi Technological University, Delhi
Email: dinesh@dtu.ac.in
Web page: http://www.dtu.ac.in/Web/Departments/InformationTechnology/faculty/dkvishwakarma.php
Biometric Research Laboratory
http://www.dtu.ac.in/Web/Departments/InformationTechnology/lab_and_infra/bml/

3/7/2023 1
 What is CNN?

3/7/2023 2
 CNN Benefits
• Equivariant activities: Replicated features do not make the neural
activities invariant to translation. The activities are equivariant.

Representation Translated
by active representation
neurons

translated
image
image

• Invariant Knowledge: If a feature is useful in some locations during

3/7/2023 4
 Possible Case

3/7/2023 5
 How CNN Works?

3/7/2023 6
 How CNN Works?

3/7/2023 7
 How CNN Works?

Gives an idea

3/7/2023 8
 How CNN Works?

First Filter

Second Filter

Third Filter

3/7/2023 9
 How CNN Works?

3/7/2023 10
 How CNN Works?

3/7/2023 11
 Convolutional Layer

3/7/2023 12
 Operations Involved in Convolution Layer
• Move filter over the image at every possible position.
• Multiply each image pixel by corresponding feature pixel.

Similarly

3/7/2023 13
 Operations Involved in Convolution Layer…
O/P of
Conv
Layer
for first
filter
• Adding all the multiplied values of filter at every move.
• Divide the sum by total number of pixels.

3/7/2023 14
 Output of Convolutional Layer

3/7/2023 15
 Activation Layer: ReLU

3/7/2023 16
 ReLU: Removes the Negative Values

3rd Feature 2nd Feature 1st Feature

3/7/2023 17
 Pooling Layer

3/7/2023 18
 Max Pooling of size: 2x2, Stride : 2

4x4
.77
7x7
3/7/2023 19
 Max Pooling of size: 2x2, Stride : 2

.77

7x7 4x4

7x7 4x4

4x4
7x7
3/7/2023 20
 Stacking of Layers
Convolutional Layer

i-th layer of CNN

3/7/2023 21
 Practical CNN Model
• Most of the practical CNN models have more number of
layers.

2x2

2x2

2x2

3/7/2023 22
 Fully Connected Layer
• Actual Classification Occurs at FC layer

Flattened the
final output
layer

3/7/2023 23
 Output

1 1
2 2
3 3
4 4
5 5
6 6
7
“1” at
7
8 8
9 9
10 10
11 11
12 12

3/7/2023 24
 Prediction

1
Now Network
2

3
is training
4
and gives
5
prediction for
6 a Test sample
7

10 Is is “X” or “O”?
11

12
Lets Compare
3/7/2023 25
 Prediction: Compare with ‘X’ and ‘O’
1 1

2 2

3 3

4 4

5 5

6 6

7 7

8 8

9 9

10 10

11 11

12 12

For ‘X’

3/7/2023 26
 Prediction: Compare with ‘X’ and ‘O’
1 1

2 2

3 3

4 4

5 5

6 6

7 7

8 8

9 9

10 10

11 11

12 12

For ‘O’

3/7/2023 27
 Result: Compare with ‘X’ and ‘O’

What is result?

Classified
as “X”

3/7/2023 28
 CNN for Edge Detection

3/7/2023 29
 Implementation of CNN
The CIFAR-
10 dataset:
This dataset
consists of
60000 32x32
colour images
in 10 classes,
with 6000
images per
class.

3/7/2023 30
 Model of CNN

3/7/2023 31
 Implementation in TensorFlow
def createModel():
model = Sequential()
# The first two layers with 32 filters of window size 3x3
model.add(Conv2D(32, (3, 3), padding='same', activation='relu', input_shape=input_shape))
model.add(Conv2D(32, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))

model.add(Conv2D(64, (3, 3), padding='same', activation='relu'))

model.add(Conv2D(64, (3, 3), padding='same', activation='relu'))

model.add(Flatten())
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(nClasses, activation='softmax'))
return model

3/7/2023 32
 References
• https://www.youtube.com/watch?v=umGJ30-15_A

3/7/2023 33