0% found this document useful (0 votes)

1 views46 pages

3 DL ConvNets

The document discusses Convolutional Neural Networks (CNNs), highlighting their advantages over fully connected layers for image processing. It covers key components such as convolution layers, pooling layers, and normalization techniques, along with various CNN architectures like AlexNet, VGG, and ResNet. Additionally, it outlines applications of CNNs in computer vision tasks such as image classification, localization, object detection, and segmentation.

Uploaded by

chakib.elmoussaoui

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

0% found this document useful (0 votes)

1 views46 pages

3 DL ConvNets

Uploaded by

chakib.elmoussaoui

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/ 46

Deep Learning

3- Convolutional Neural Networks

Prof. Monir EL ANNAS

Deep Learning /3- Convolutional Neural Networks 1

Convolutional Neural Networks

Problems with Fully Connected Layers on Images

Deep Learning /3- Convolutional Neural Networks 2

Convolutional Neural Networks
Problems with Fully Connected Layers on Images

Deep Learning /3- Convolutional Neural Networks 3

Convolutional Neural Networks
Problems with Fully Connected Layers on Images

Deep Learning /3- Convolutional Neural Networks 4

Convolutional Neural Networks

Problems with Fully Connected Layers on Images

Deep Learning /3- Convolutional Neural Networks 5

Convolutional Neural Networks

Problems with Fully Connected Layers on Images

Deep Learning /3- Convolutional Neural Networks 6

Convolutional Neural Networks

Problems with Fully Connected Layers on Images

Deep Learning /3- Convolutional Neural Networks 7

Convolutional Neural Networks
Problems with Fully Connected Layers on Images

Deep Learning /3- Convolutional Neural Networks 8

Convolutional Neural Networks
Problems with Fully Connected Layers on Images

Deep Learning /3- Convolutional Neural Networks 9

Convolutional Neural Networks

Problems with Fully Connected Layers on Images

Deep Learning /3- Convolutional Neural Networks 10

Convolutional Neural Networks

Convolution Layer

Deep Learning /3- Convolutional Neural Networks 11

Convolutional Neural Networks

Convolution Layer

Deep Learning /3- Convolutional Neural Networks 12

Convolutional Neural Networks

Convolution Layer

Deep Learning /3- Convolutional Neural Networks 13

Convolutional Neural Networks

Convolution Layer

Deep Learning /3- Convolutional Neural Networks 14

Convolutional Neural Networks

Dimensions of a Convolution Layer: Stride

Deep Learning /3- Convolutional Neural Networks 15

Convolutional Neural Networks

Dimensions of a Convolution Layer: Stride

Deep Learning /3- Convolutional Neural Networks 16

Convolutional Neural Networks

Dimensions of a Convolution Layer: Stride

Deep Learning /3- Convolutional Neural Networks 17

Convolutional Neural Networks

Dimensions of a Convolution Layer: Stride

Deep Learning /3- Convolutional Neural Networks 18

Convolutional Neural Networks

Dimensions of a Convolution Layer: Padding

Deep Learning /3- Convolutional Neural Networks 19

Convolutional Neural Networks

Dimensions of a Convolution Layer: Example

Deep Learning /3- Convolutional Neural Networks 20

Convolutional Neural Networks

Dimensions of a Convolution Layer: Example

Deep Learning /3- Convolutional Neural Networks 21

Convolutional Neural Networks
Pooling Layer

Deep Learning /3- Convolutional Neural Networks 22

Convolutional Neural Networks
Pooling Layer

Deep Learning /3- Convolutional Neural Networks 23

Convolutional Neural Networks
Normalization for ConvNets

Spacial Bach Normalization for

Bach Normalization for FC networks
ConvNets
• Input size (N, D)
• Input size (N, C , W , H)
• Compute minibatch mean and
• Compute minibatch mean and
variance across N (i.e., compute
variance across N, W , H (i.e.,
mean/variance for each feature
compute mean/variance for each
dimension)
channel C )
x :N ×D
x :N ×C ×H ×W
µ, σ : 1 × D
µ, σ : 1 × C × 1 × 1
γ, β : 1 × D
γ, β : 1 × C × 1 × 1
(x − µ)
y =γ +β (x − µ)
σ y =γ +β
σ

Deep Learning /3- Convolutional Neural Networks 24

Convolutional Neural Networks

Final Fully Connected Layer

Deep Learning /3- Convolutional Neural Networks 25

Convolutional Neural Networks

The ConvNet Building Blocks

Deep Learning /3- Convolutional Neural Networks 26

Convolutional Neural Networks

ConvNets Architectures

Deep Learning /3- Convolutional Neural Networks 27

Convolutional Neural Networks

ConvNets Architectures

Deep Learning /3- Convolutional Neural Networks 28

Convolutional Neural Networks
ConvNets Architectures: AlexNet

Deep Learning /3- Convolutional Neural Networks 29

Convolutional Neural Networks

ConvNets Architectures: AlexNet

Deep Learning /3- Convolutional Neural Networks 30

Convolutional Neural Networks

ConvNets Architectures: VGG

Deep Learning /3- Convolutional Neural Networks 31

Convolutional Neural Networks
ConvNets Architectures: VGG

Deep Learning /3- Convolutional Neural Networks 32

Convolutional Neural Networks

ConvNets Architectures: GoogLeNet

Deep Learning /3- Convolutional Neural Networks 33

Convolutional Neural Networks
ConvNets Architectures: GoogLeNet

Deep Learning /3- Convolutional Neural Networks 34

Convolutional Neural Networks

ConvNets Architectures: ResNet

Deep Learning /3- Convolutional Neural Networks 35

Convolutional Neural Networks
ConvNets Architectures: ResNet

Deep Learning /3- Convolutional Neural Networks 36

Convolutional Neural Networks

ConvNets Architectures: EfficientNet

Deep Learning /3- Convolutional Neural Networks 37

Convolutional Neural Networks

ConvNets Architectures

Network Year Key Feature Top-5 Accuracy Parameters FLOPs

AlexNet 2012 Deeper 84.70% 62M 1.5B
VGGNet 2014 Fixed-size kernels 92.30% 138M 19.6B
Inception 2014 Wider – Parallel kernels 93.30% 6.4M 2B
ResNet-152 2015 Shortcut connections 95.51% 60.3M 11B
EfficientNet-B7 2019 Compound scaling 97.10% 66M 37B

Deep Learning /3- Convolutional Neural Networks 38

Convolutional Neural Networks

ConvNets Architectures: Transfer Learning

Deep Learning /3- Convolutional Neural Networks 39

Convolutional Neural Networks

ConvNets Architectures: Summary

• Design your network according to your task and resources

• Take into consideration
• # Parameters
• # FLOPs
• Memory Size
• Use existing architectures when possible
• Use pre-trained models when possible

Deep Learning /3- Convolutional Neural Networks 40

Convolutional Neural Networks

Applications: Typical Computer Vision Tasks

Deep Learning /3- Convolutional Neural Networks 41

Convolutional Neural Networks

Applications: CNNs for Image Classification

• Backend: Convolution (Conv) + Pooling (Pool)

• Frontend: Fully Connected (FC) or Global Average Pooling (GAP) + Softmax

Deep Learning /3- Convolutional Neural Networks 42

Convolutional Neural Networks

Applications: CNNs for Image Localization

• Backend: Conv + Pool

• Frontend:
• One head for classification
• One head for bounding box regression

Deep Learning /3- Convolutional Neural Networks 43

Convolutional Neural Networks

Applications; CNNs for Object Detection

• Two stages:
• Region proposal
• Proposal classification + refinement
• R-CNN
• External mechanism for proposal
generation
• Fast R-CNN
• Learnt proposals, two networks
• Faster R-CNN
• Shared backbone for both stages

Deep Learning /3- Convolutional Neural Networks 44

Convolutional Neural Networks

Applications: CNNs for Semantic Segmentation

Deep Learning /3- Convolutional Neural Networks 45

Convolutional Neural Networks

Applications: CNNs for Instance Segmentation

• Mask R-CNN
• Faster R-CNN + Additional head for instance segmentation

Deep Learning /3- Convolutional Neural Networks 46

Neural Network Notes
No ratings yet
Neural Network Notes
268 pages
Dlai DL CNN
No ratings yet
Dlai DL CNN
193 pages
Lec 03
No ratings yet
Lec 03
40 pages
ML Seminar
No ratings yet
ML Seminar
58 pages
Deep Learning Convolution Neural Networks
No ratings yet
Deep Learning Convolution Neural Networks
73 pages
Hot Chips Overview
No ratings yet
Hot Chips Overview
47 pages
Lecture # 4-1 Convolutional Neural Networks
No ratings yet
Lecture # 4-1 Convolutional Neural Networks
62 pages
CNN and Autoencoder
No ratings yet
CNN and Autoencoder
56 pages
06 CNN Convolutional Neural Network
No ratings yet
06 CNN Convolutional Neural Network
133 pages
DL Unit-Ii
No ratings yet
DL Unit-Ii
34 pages
FT04 Haghighat Independent 2023
No ratings yet
FT04 Haghighat Independent 2023
40 pages
20CS4701A
No ratings yet
20CS4701A
2 pages
Unit 3
No ratings yet
Unit 3
105 pages
Images and Convolutional Neural Networks: Practical Deep Learning
No ratings yet
Images and Convolutional Neural Networks: Practical Deep Learning
34 pages
CNN Short
No ratings yet
CNN Short
61 pages
Kernel Slides
No ratings yet
Kernel Slides
33 pages
Unit III
No ratings yet
Unit III
89 pages
Introduction To Deep Learning
No ratings yet
Introduction To Deep Learning
47 pages
Convolutional Neural Networks - Deeplearning-Notes
No ratings yet
Convolutional Neural Networks - Deeplearning-Notes
43 pages
DL Unit3 1
No ratings yet
DL Unit3 1
67 pages
03 CNN
No ratings yet
03 CNN
33 pages
Anthony
No ratings yet
Anthony
33 pages
Ch-3 Convolutional Neural Networks (CNNS)
No ratings yet
Ch-3 Convolutional Neural Networks (CNNS)
11 pages
Unit III
No ratings yet
Unit III
89 pages
Convolutional Neural Networks - Annotated
No ratings yet
Convolutional Neural Networks - Annotated
83 pages
L3 - UUCLxDeepMind DL2020
No ratings yet
L3 - UUCLxDeepMind DL2020
110 pages
Lec 2
No ratings yet
Lec 2
42 pages
CS601 Machine Learning Unit 3
No ratings yet
CS601 Machine Learning Unit 3
47 pages
Additional CNN
No ratings yet
Additional CNN
82 pages
Convolutional Neural Network Architecture - CNN Architecture
No ratings yet
Convolutional Neural Network Architecture - CNN Architecture
13 pages
4a Convolutional Neural Networks
No ratings yet
4a Convolutional Neural Networks
56 pages
Unit Iii Convolutional Networks and Sequence Modelling
No ratings yet
Unit Iii Convolutional Networks and Sequence Modelling
38 pages
Deep Learning (MODULE-3)
No ratings yet
Deep Learning (MODULE-3)
85 pages
ch4 CNN
No ratings yet
ch4 CNN
35 pages
Image Classification Using Convolutional Neural Networks (CNNS)
No ratings yet
Image Classification Using Convolutional Neural Networks (CNNS)
61 pages
5-Convolutional Neural Network
No ratings yet
5-Convolutional Neural Network
43 pages
CVlecture 5
No ratings yet
CVlecture 5
56 pages
AE556 2024 Topic4 CNN
No ratings yet
AE556 2024 Topic4 CNN
26 pages
Deep Representation Learning Techniques For Audio Signal Processing
No ratings yet
Deep Representation Learning Techniques For Audio Signal Processing
152 pages
DL Inference FPGA Class1
No ratings yet
DL Inference FPGA Class1
56 pages
Convolutional Neural Networks (CNN) : Convolutions
No ratings yet
Convolutional Neural Networks (CNN) : Convolutions
17 pages
Cnnbasics 171028092801
No ratings yet
Cnnbasics 171028092801
43 pages
Unit IV Deep Leraning
No ratings yet
Unit IV Deep Leraning
35 pages
Convolutional Neural Networks: CS 535 Deep Learning, Winter 2020 Fuxin Li
No ratings yet
Convolutional Neural Networks: CS 535 Deep Learning, Winter 2020 Fuxin Li
44 pages
L10 - Intro - To - Deep - Learning
No ratings yet
L10 - Intro - To - Deep - Learning
75 pages
Project Exhibition 2
No ratings yet
Project Exhibition 2
42 pages
COMP3220 Lect 11 - Introduction To Convolutional Neural Networks
No ratings yet
COMP3220 Lect 11 - Introduction To Convolutional Neural Networks
13 pages
UNIT-III DeepLearning Notes
No ratings yet
UNIT-III DeepLearning Notes
30 pages
Chapter 5 Deep Learning
No ratings yet
Chapter 5 Deep Learning
35 pages
Super VIP Cheatsheet - Deep Learning
No ratings yet
Super VIP Cheatsheet - Deep Learning
47 pages
Module 3 Notes
No ratings yet
Module 3 Notes
22 pages
12-DL-Deep Learning For GANS
No ratings yet
12-DL-Deep Learning For GANS
75 pages
Typical CNN (Convolutional Neural Network) Architecture: CHARAN S (1VE20CA005) Cse-Ai, Svce
No ratings yet
Typical CNN (Convolutional Neural Network) Architecture: CHARAN S (1VE20CA005) Cse-Ai, Svce
13 pages
07 Ais302 CNN
No ratings yet
07 Ais302 CNN
56 pages
Convolutional Neural Networks Notes
No ratings yet
Convolutional Neural Networks Notes
29 pages
CNN Architecture
No ratings yet
CNN Architecture
24 pages
An Overview of Convolutional Neural Network Architectures For Deep Learning
No ratings yet
An Overview of Convolutional Neural Network Architectures For Deep Learning
22 pages
Convolutional Neural Network - Towards Data Science PDF
No ratings yet
Convolutional Neural Network - Towards Data Science PDF
10 pages
Deep Learning Notes For Easy Access
No ratings yet
Deep Learning Notes For Easy Access
14 pages
DL Question Bank
No ratings yet
DL Question Bank
23 pages
Andrej Karpathy Blog Karpathhy
No ratings yet
Andrej Karpathy Blog Karpathhy
4 pages
Deep LearningUNIT-IV
No ratings yet
Deep LearningUNIT-IV
16 pages
11.RNN and Transformers
No ratings yet
11.RNN and Transformers
100 pages
2023 Automated Fish Classification Using
No ratings yet
2023 Automated Fish Classification Using
19 pages
Introduction To Neural Network
No ratings yet
Introduction To Neural Network
17 pages
Neural Network Architecture
No ratings yet
Neural Network Architecture
3 pages
Unit 5b - Natural Language Processing
No ratings yet
Unit 5b - Natural Language Processing
41 pages
Introduction To Convolution Neural Network
No ratings yet
Introduction To Convolution Neural Network
6 pages
DL 4
No ratings yet
DL 4
11 pages
Master Spilak Bruno
No ratings yet
Master Spilak Bruno
73 pages
CISC 867 Deep Learning: 12. Recurrent Neural Networks
No ratings yet
CISC 867 Deep Learning: 12. Recurrent Neural Networks
72 pages
Introduction To ANN Set 4 (Network Architectures)
No ratings yet
Introduction To ANN Set 4 (Network Architectures)
6 pages
Rosen Blatt's Perceptron Model
No ratings yet
Rosen Blatt's Perceptron Model
11 pages
In-The-Wild Deepfake Detection Using Adaptable CNN Models With Visual Class Activation Mapping For Improved Accuracy
No ratings yet
In-The-Wild Deepfake Detection Using Adaptable CNN Models With Visual Class Activation Mapping For Improved Accuracy
6 pages
Lecture 11 Advanced CNN
No ratings yet
Lecture 11 Advanced CNN
42 pages
Mathematics of Deep Learning 1687444204
No ratings yet
Mathematics of Deep Learning 1687444204
45 pages
Python Deep Learning Tutorial
0% (1)
Python Deep Learning Tutorial
17 pages
Pa TH MDM
No ratings yet
Pa TH MDM
4 pages
Basics of DL: Prof. Leal-Taixé and Prof. Niessner 1
No ratings yet
Basics of DL: Prof. Leal-Taixé and Prof. Niessner 1
76 pages
PVSNet Palm Vein Authentication
No ratings yet
PVSNet Palm Vein Authentication
8 pages
Classification of Iris Data Set PDF
No ratings yet
Classification of Iris Data Set PDF
21 pages
DL Assignment 5
No ratings yet
DL Assignment 5
4 pages
Question Bank New
No ratings yet
Question Bank New
3 pages
Adobe Scan Dec 17, 2023
No ratings yet
Adobe Scan Dec 17, 2023
1 page
hw1 Deep Learning SoSe2024
No ratings yet
hw1 Deep Learning SoSe2024
2 pages
Generative Adversarial Networks: (Ian Goodfellow)
No ratings yet
Generative Adversarial Networks: (Ian Goodfellow)
3 pages
CS485 Ch5 Transformers
No ratings yet
CS485 Ch5 Transformers
50 pages
Techniques and Tools for Artificial Intelligence. Neural Networks via R and PYTHON
From Everand
Techniques and Tools for Artificial Intelligence. Neural Networks via R and PYTHON
César Pérez López
No ratings yet
Convolutional Neural Networks in Python: Beginner's Guide to Convolutional Neural Networks in Python
From Everand
Convolutional Neural Networks in Python: Beginner's Guide to Convolutional Neural Networks in Python
Frank Millstein
No ratings yet
Convolutional Neural Networks: Fundamentals and Applications for Analyzing Visual Imagery
From Everand
Convolutional Neural Networks: Fundamentals and Applications for Analyzing Visual Imagery
Fouad Sabry
No ratings yet

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.