Scribd
Upload
21 views39 pages

LeNet

Uploaded by

nericjoel2003
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
21 views39 pages

LeNet

Uploaded by

nericjoel2003
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 39

20IS712

Deep Learning
300 (3)
Dr. Lekshmi R. R.
Asst. Prof.
Department of Electrical & Electronics
Engineering
Amrita School of Engineering 1
LeNet
Neural Network
• LeNet – first convolutional network
• Used to classify 2D images
– Grey scale images
– During LeNet development
– No GPU and slow CPU to train LeNet
– LeNet – trained used grey scale
– 32x32x1
– Height: 32
– Width: 32
– No. of channel: 1 (Grey scale)
Architecture

Step 1 (C1) Convolution Layer 2

• Input - 32х32 grey scale image


• Input goes to 1st convolution layer
• Filter size : 5х5
• Striding: 1
Suggested by LeCun et.al to detect signature in mail • No. of feature maps: 6

Character is recognized through multiple layers


32

32
5

5
28
28
Architecture

Step 1 Convolution Layer 2

• Input - 32х32 grey scale image


• Input goes to 1st convolution layer
• Filter size : 5х5
Suggested by LeCun et.al to detect signature in mail • Striding: 1
• No. of feature maps: 6
• Output: 28х28 with 6 feature
Character is recognized through multiple layers maps (reduced)
Architecture

Step 2 (C2)

• Down sample
– Average pooling
• Input: 28x28
• Filter size : 2х2
• Striding : 2
28

28
2
2
14
14
Architecture

Step 2

• Down sample
– Average pooling
• Input: 28x28
• Filter size : 2х2
• Striding : 2
• Output : 14х14 with 6 feature
maps
Architecture

Step 3 (C3) Convolution Layer 2

• Input - 14х14 image


• Filter size : 5х5
• Striding: 1
• No. of feature maps: 16
14

14
5

5
10
10
Architecture

Step 3 Convolution Layer 2

• Input - 14х14 image


• Filter size : 5х5
• Striding: 1
• No. of feature maps: 16
• Output: 10х10 with 16 feature
maps (reduced)
Architecture

Step 4

• Down sample
– Average pooling
• Input: 10x10
• Filter size : 2х2
• Striding : 2
10

10
2

2
5
5
Architecture

Step 4

• Down sample
– Average pooling
• Input: 10x10
• Filter size : 2х2
• Striding : 2
• Output : 5х5 with 16 feature maps
Architecture

Step 5 Convolution layer

• Input - 5х5 image


• Fully connected layer
• No. of feature maps: 120
• Size : 1x1
• Each 120 units connected to 400
nodes of previous layer (16x5x5)
Architecture

Step 6 (C6) Convolution layer

• Fully connected layer


• No. of feature maps: 84
Architecture

Step 7 (C7)

• Fully connected layer


• Softmax activation function
output layer
– With 10 possible values
• 0-9
Thank you

You might also like