0% found this document useful (0 votes)
15 views9 pages

Convolution Over Volumes

Uploaded by

shubhodippal01
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)
15 views9 pages

Convolution Over Volumes

Uploaded by

shubhodippal01
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/ 9

Convolution over volumes

Dr. Chiradeep Mukherjee


Department of CST & CSIT
Convolution in RGB (3D) Image

What will be the


size of Feature
Map?

3X3 Kernel How it will be


created?
6X6 Image Space
Convolution in RGB (3D) Image
25 2 10 0 27-3 75 24 25
20 2 9 0 58-3 25 14 25 2 0 -3
211 21 0 21-3 52 20 25 2 0 -3
27-3 2275 22
25 2 10 0 22 24 65
25 0 25 1 0 -3
20 2 9 58-3 2525 25
25
0
85 2 25
14 25
21-3 15
211 21 0 15 52 15 25 25 25
20 25
25 2 10 0 27-3 2275 2224 25 65 0 25
22
20 2 9 0 58-3 2525 2514 25 85 2 25
211 21 0 21-3 1552 1520 15 25 25 25
25
22 22 22 65 0 25
How many weights we need
25 25 25 85 2 25 to define?
15 15 15 25 25 25
Convolution in RGB (3D) Image
Vertical Edge

4X4X2
Feature Map
How many weights we need
to define?: 3X3X3X2 Horizontal Edge
k is filter
number

Summary: nXnXnc fXfXfc (n-f+1) X (n-f+1) X k


Understanding the dimensions clearly

Dim: 3X3X2

Dim:
10X10X5
Dim: 5X5X3
Calculations of the dimensions for a traditional ConvNet
Weight: 3X3X3X10 Weight: 5X5X10X20

Filter: 3X3X3 Filter: 5X5X10


Dim: Stride=1 Stride=2
39X39X3 Padding=0
Dim: Padding=0
Dim:
10 filters 20 filters
37X37X10 17X17X20
Weight: 5X5X20X40
Vectorization
Class 1 Filter: 5X5X20
Stride=2
Class 2 Dim: Dim: Padding=0
1960X1 40 filters
7X7X40
Number of parameters in a traditional CNN
Weight: 3X3X3X10 Weight: 5X5X10X20

Total: 270 Total: 5000

Weight: 5X5X20X40

Vectorization
Class 1 Total: 20000

Class 2 Dim:
1960X1

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy