0% found this document useful (0 votes)
10 views11 pages

Inception Practice: CNN Layer Breakdown

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
0% found this document useful (0 votes)
10 views11 pages

Inception Practice: CNN Layer Breakdown

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

Inception Practice

INSTRUCTOR NAME: SHUKDEV DATTA


ML DEVELOPER AT INNOVATIVE SKILLS
Inception Practice
Inception Practice

1x1 CONV
Input image = 28x28x256
Filter = 1x1x256
#F = 64
S=1
P=0
Output = floor[(28-1/1)+1]x floor[(28-1/1)+1]
= 28x28x64

No of parameters = (1x1x256 +1)*64


Inception Practice

1x1 CONV
Input image = 28x28x256
Filter = 1x1x256
#F = 64
S=1
P=0
Output = floor[(28-1/1)+1]x floor[(28-1/1)+1]
= 28x28x64

No of parameters = (1x1x256 +1)*64


Inception Practice

3x3 POOL
Input image = 28x28x256
Filter = 3x3
S=1
P=1
Output = floor[(28-3+2/1)+1]x floor[(28-
3+2/1)+1]
= 28x28x256

No of parameters = 0
Inception Practice

1x1 CONV
Input image = 28x28x256
Filter = 1x1x256
#F = 128
S=1
P=0
Output = floor[(28-1/1)+1]x floor[(28-1/1)+1]
= 28x28x128

No of parameters = (1x1x256 +1)*128


Inception Practice

3x3 CONV
Input image = 28x28x64
Filter = 3x3x64
#F = 192
S=1
P = 21
Output = floor[(28-3+2/1)+1]x floor[(28-
3+2/1)+1]
= 28x28x192

No of parameters = (3x3x64+1)*192
Inception Practice

5x5 CONV
Input image = 28x28x64
Filter = 5x5x64
#F = 96
S=1
P=2
Output = floor[(28-5+2*2/1)+1]x floor[(28-
5+2*2/1)+1]
= 28x28x96

No of parameters = (5x5x64 +1)*96


Inception Practice

1x1 CONV
Input image = 28x28x256
Filter = 1x1x256
#F = 64
S=1
P=0
Output = floor[(28-1/1)+1]x floor[(28-1/1)+1]
= 28x28x64

No of parameters = (1x1x256+1)*64
Inception Practice

Filter Concatenation

28x28x128+28x28x192+28x28x96+28x28x64
=28x28x(128+192+96+64)
= 28x28x480
Thank You!!!

You might also like