Scribd
Upload
23 views21 pages

CNN3 Pooling and Fully Contected Layers

Uploaded by

Zarnain Sheikh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
23 views21 pages

CNN3 Pooling and Fully Contected Layers

Uploaded by

Zarnain Sheikh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 21

Pooling and Fully Connected Layers

Pooling Layer

Reduces computational
224x224x64
complexity
112x112x64
Combats overfitting

Encourages translational
invariance
pool
Pooling Layer
single depth slice Pooling layers also have
1 1 2 88 filters with width and stride

max pool 2x2 filter


5 66 7 4 and stride 2

3 2 1 0
1 2 3 44
Pooling Layer
single depth slice Pooling layers also have
1 1 2 88 filters with width and stride

max pool 2x2 filter


5 66 7 4 and stride 2

3 2 1 0
1 2 3 44
Pooling Layer
single depth slice Pooling layers also have
1 1 2 88 filters with width and stride

max pool 2x2 filter


5 66 7 4 and stride 2

3 2 1 0
1 2 3 44
Pooling Layer
single depth slice Pooling layers also have
1 1 2 88 filters with width and stride

max pool 2x2 filter


5 66 7 4 and stride 2

3 2 1 0
1 2 3 4
Pooling Layer
single depth slice Pooling layers also have
1 1 2 8 filters with width and stride

max pool 2x2 filter


5 66 7 4 and stride 2

3 2 1 0
1 2 3 4
Pooling Layer
single depth slice Pooling layers also have
1 1 2 8 filters with width and stride

max pool 2x2 filter


5 66 7 4 and stride 2

3 2 1 0
1 2 3 44
Pooling Layer
single depth slice Pooling layers also have
1 1 2 8 filters with width and stride

max pool 2x2 filter


5 6 7 4 and stride 2 6
3 2 1 0
1 2 3 44
Pooling Layer
single depth slice Pooling layers also have
1 1 2 8 filters with width and stride

max pool 2x2 filter


5 6 7 4 and stride 2 6
3 2 1 0
1 2 3 44
Pooling Layer
single depth slice Pooling layers also have
1 1 2 8 filters with width and stride

max pool 2x2 filter


5 6 7 4 and stride 2 6 8
3 2 1 0
1 2 3 44
Pooling Layer
single depth slice Pooling layers also have
1 1 2 8 filters with width and stride

max pool 2x2 filter


5 6 7 4 and stride 2 6 8
3 2 1 0 3 4
1 2 3 4
Pooling Layer
single depth slice Pooling layers also have
1 1 2 8 filters with width and stride

max pool 2x2 filter Pooling typically uses the max;


5 6 7 4 and stride 2 6 8 could also use the mean

3 2 1 0 3 4
1 2 3 4
Pooling Layer
single depth slice Pooling layers also have

16 1 2 8 filters with width and stride

max pool 2x2 filter Pooling typically uses the max;


5 6 7 4 and stride 2 6 8 could also use the mean

3 2 1 0 3 4
1 2 3 4
Pooling Layer
single depth slice Pooling layers also have

16 1 2 8 filters with width and stride

max pool 2x2 filter Pooling typically uses the max;


5 6 7 4 and stride 2 6 8 could also use the mean

3 2 1 0 3 4 Max pooling picks out strong


activation with some position
1 2 3 4 independence
Fully Connected
Layer
input convolutional layer pooling layer convolutional layer pooling layer fully connected
28x28 24x24 12x12 8x8 4x4 readout
10 filters
10 filters 0
20 filters 1
20 filters
2
3
4
5
6
convolution max pooling convolution max pooling 7
(5x5 kernel) (2x2) (5x5 kernel) (2x2) 8
9
Fully Connected
Layer fully connected
pooling layer readout
4x4
0
20 filters
1

9
Fully Connected
Layer 320x1 fully connected
pooling layer readout
4x4
0
20 filters
1

9
Fully Connected
Layer 320x1 fully connected
pooling layer readout
4x4
0
20 filters
1

9
Fully Connected
Layer 320x1 fully connected fully connected
pooling layer (latent) readout
4x4
0
20 filters
1

9
Fully Connected
Layer 320x1 fully connected fully connected
pooling layer (latent) readout
4x4
0
20 filters
1

You might also like