You are on page 1of 16

|| Jai Sri Gurudev ||

SJB Institute of Technology


#67, BGS Health & Education City, Dr. Vishnuvardhan Road, Kengeri, 
       Bengaluru – 560060, KARNATAKA, INDIA

Technical Seminar –01


Generative Adversarial Network ( GAN )
Presented By

Student’s name       [USN] Under the guidance of


Rhythm Bhatnagar 1JB18CS116 Dr. Krishna A N
Head of Department
Computer Science and Engineering
S J B INSTITUTE OF TECHNOLOGY
TABLE OF CONTENTS
 INTRODUCTION

 LITERATURE SURVEY

 MOTIVATION AND PROBLEM STATEMENT

 BASICS OF GAN

 TRAINING OF GAN

 APPLICATION OF GAN

 DRAWBACKS OF GAN
INTRODUCTION

• A generative adversarial network (GAN) is a machine learning model in which two neural networks
 compete with each other to become more accurate in their predictions.
• The two neural networks that make up a GAN are referred to as the Generator and the Discriminator.
• The goal of the generator is to artificially manufacture outputs that could easily be mistaken for real data.
The goal of the discriminator is to identify which outputs it receives have been artificially created.
• The generator will begin to produce higher-quality output and the discriminator will become better at
flagging data that has been artificially created.
LITERATURE SURVEY
Author Title Year Technique Learning

Zhao Fan, Jin Hu Review and Prospect of 2019 GAN and its applications About Generative
Research on Generative adversarial Model and its
Adversarial Networks applications

Shailender Kumar, Sumit A Detailed Study on 2020 Generative Adversarial A Detailed Study on
Dhawan Generative Adversarial Networks Generative Adversarial
Networks Networks and its Types.

Liang Gonog1,2 and A Review: Generative 2019 Generative Adversarial Basics of GAN, Training of
Yimin Zhou1 Adversarial Networks Networks GAN, Usage of the GANs
MOTIVATION AND PROBLEM STATEMENT

PROBLEM STATEMENT: GAN (GENERATIVE ADVERSARAL NETWORK) , its advantages and drawbacks in
practical applications.
MOTIVATION: Person Re-identification is an application in which a person/ multiple people are re identified
either through the same camera or through multiple cameras.
GAN is one of the few techniques that is used in Person Re-Identification to achieve efficient and fast
results.
BASICS OF GAN

• GAN stands for Generative Adversarial Network


• GAN are deep neural network architecture that comprises of two neural networks that competes
against each other, hence the term “adversarial”.
• Aim: To generate data that mimics some true distribution data.
• Two classes of models used in GAN are:
1. Discriminative Model: Discriminates between two classes of data.
2. Generative Model: A generative model G to be trained on training data X sampled from some True
distribution D, given some standard random distribution Z, The Generative
Model produces a distribution D’ which is close to D.
PICTORIAL REPRESENTATION

Here,
• Z is random distribution. z represents sample present in Z.
• D is some True Distribution
• G is the Generative model which is a CNN.
• Objective of G is to convert Z into D’ which is as close as possible to D.
• Discriminator discriminates between two classes and classify by producing a value 0(Fake) or 1(Not Fake).
• Generator generates Fake Samples D’ which are as close as possible to real samples D.
• Finetuning is done by adjusting the weights and bias of the generative and discriminative models.
• Finetuning of generator is done one by one till discriminator produces a value of 0.5
i.e. it fails to classify whether the image is real or fake.
• Generator G produces G(z) which is fed to the Discriminator.
• A sample from G(z) (y=0) and D (y=1) is taken and fed to the Discriminator which is expected
to classify them correctly.
• After which loss function L is calculated and we backpropagate the errors.
• These Backpropagated errors adjusts the weights of the Generator.
• This process is repeated for many iterations till the Generator produces a sample D’ which
fools the Discriminator.
• For Generator the function log(1-D(G(z))) is considered.
• For Discriminator the function { log(D(x)) + log(1-D(G(z))) } is considered
• The objective is to minimize function of G and maximize the function of D
• Function value of G is minimized by taking D(G(z)) to maxima point, i.e. D(G(z)) produces a value of 1.
APPLICATION OF GAN:
1. In General Model Of Person Re-id.
2. Pose Guided Person Generation
• The design composes of a 2-stage image generator and a discriminator. The generator reconstruct an
image using the meta-data (pose) and the original image. The discriminator uses the original image
as part of the label input to a CGAN design.
3. Cycle GAN
Cross-domain transfer GANs will be likely the first batch of commercial applications. These GANs transform
images from one domain (say Horses) to another domain (Zebra).
4. Image Inpainting
Repair images have been an important subject decades ago. GAN is used to repair images and fill the missing part
with created “content”.
DRAWBACKS OF GAN

Disadvantages of Generative Adversarial Networks (GAN’s):


• Harder to train:  You need to provide different types of data continuously to check if it works
accurately or not.
• Generating results from text or speech is very complex.
THANK YOU

You might also like