Knowledge base: Warsaw University of Technology

Settings and your account

Back

Extending a dataset using Genarative Adversarial Networks (GANs)

Assam Akhtar Chaudhary

Abstract

The thesis presents a technique to augment a dataset by synthesizing images using Generative Adversarial Networks (GANs). I have used two training datasets; the MNIST dataset consisting of handwritten digits from 0-9 and a dataset of fetal ultrasound images containing only 360 images. I investigate different architectures of GANs to model the underlying distribution of training data to allow for additional synthetic data to be sampled and used to augment the given real dataset.
Diploma type
Engineer's / Bachelor of Science
Diploma type
Engineer's thesis
Author
Assam Akhtar Chaudhary (FEIT/ICS) Assam Akhtar Chaudhary,, The Institute of Computer Science (FEIT/ICS)Faculty of Electronics and Information Technology (FEIT)
Title in Polish
Rozszerzenie danych za pomocą GANs
Supervisor
Robert Marek Nowak (FEIT/ICS) Robert Marek Nowak,, The Institute of Computer Science (FEIT/ICS)Faculty of Electronics and Information Technology (FEIT)
Certifying unit
Faculty of Electronics and Information Technology (FEIT)
Affiliation unit
The Institute of Computer Science (FEIT/ICS)
Study subject / specialization
, Informatyka (Computer Science)
Language
(en) English
Status
Finished
Defense Date
26-02-2019
Issue date (year)
2019
Internal identifier
45/19 (2638)
Reviewers
Robert Marek Nowak (FEIT/ICS) Robert Marek Nowak,, The Institute of Computer Science (FEIT/ICS)Faculty of Electronics and Information Technology (FEIT) Paweł Zawistowski (FEIT/ICS) Paweł Zawistowski,, The Institute of Computer Science (FEIT/ICS)Faculty of Electronics and Information Technology (FEIT)
Keywords in Polish
Generative Adversarial Networks, rozszerzenie danych,Głębokie sieci neuronowe.
Keywords in English
Generative Adversarial Networks, Dataset augmentation, Deep Convolutional NeuralNetworks.
Abstract in Polish
W tym artykule przedstawię technikę rozszerzenie (augmentacji) danych, która syntetyzuje obrazy przy użyciu sieci neuronowych GAN (Generative Adversarial Networks). Użyłem dwóch zestawów danych trenujących; zestaw danych MNIST dla odręcznych cyfr od 0 do 9 oraz zestaw danych z obrazów USG płodu zawierających jedynie obrazy 360 obrazów. Badam różne architektury GAN, aby modelować podstawowy rozkład danych treningowych, aby umożliwić pobieranie próbek danych syntetycznych i ich wykorzystanie do rozszerzenia danego rzeczywistego zestawu danych.
File
  • File: 1
    Thesis.pdf
Request a WCAG compliant version
Local fields
Identyfikator pracy APD: 32145

Uniform Resource Identifier
https://repo.pw.edu.pl/info/bachelor/WUT3a6d87538d2640b0944946e9b78067e7/
URN
urn:pw-repo:WUT3a6d87538d2640b0944946e9b78067e7

Confirmation
Are you sure?
Report incorrect data on this page