Translasi Domain Citra Menggunakan Metode Unpaired Image-to-image Translation CycleGAN

Tjahjono, Benedictus Kenny (2024) Translasi Domain Citra Menggunakan Metode Unpaired Image-to-image Translation CycleGAN. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003201159-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 July 2026. Download (5MB) | Request a copy

Abstract

Dalam era globalisasi dan digitalisasi saat ini, teknologi berperan besar dalam berbagai sektor, termasuk sektor pariwisata. Pengembangan teknologi dalam bidang computer vision, yaitu klasifikasi dan image-to-image translation, dapat digunakan untuk mengklasifikasikan objek pariwisata dan mengubah domain foto sehingga memungkinkan pengguna untuk mendapatkan foto yang sesuai ekspektasi dan keinginan mereka. Klasifikasi pada penelitian ini menggunakan tiga jenis arsitektur pre-trained model Convolutional Neural Network (CNN) yaitu VGG16, GoogLeNet, dan ResNet serta tiga jenis optimizer, yaitu SGD, RMSProp, dan Adam. Preprocessing citra yang dilakukan meliputi rescale berukuran 128 untuk klasifikasi dan berukuran 256 untuk translasi serta normalisasi nilai piksel setiap citra. Model pre-trained CNN terbaik pada penelitian ini adalah VGG16 dengan menggunakan Adam optimizer dengan learning rate sebesar 0,001. Implementasi image-to-image translation dengan metode CycleGAN dalam melakukan translasi domain citra pantai dan hutan memberikan hasil yang memuaskan. Generator dan diskriminator yang sudah dilatih untuk translasi domain citra dapat memberikan citra hasil bangkitan yang realistis dan nampak natural meskipun terdapat kekurangan pada masing-masing generator.
==================================================================================================================================
In the current era of globalization and digitalization, technology plays a major role in various sectors, including the tourism sector. Technology development in the field of computer vision, namely classification and image-to-image translation, can be used to classify tourism objects and change the domain of photos so as to enable users to get photos that meet their expectations and desires. The classification in this study uses three types of pre-trained architecture of Convolutional Neural Network (CNN) models, namely VGG16, GoogLeNet, and ResNet as well as three types of optimizers, namely SGD, RMSProp, and Adam. Image preprocessing includes rescale of 128 for classification and 256 for translation and normalization of the pixel value of each image. The best CNN pre-trained model in this study is VGG16 using Adam optimizer with a learning rate of 0.001. The generators and discriminators that have been trained for image domain translation can provide realistic and natural-looking generated images despite the shortcomings of each generator.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	CycleGAN, Image-to-image translation, Klasifikasi, Objek Pariwisata, Classification; CycleGAN, Image-to-image translation, Tourism Object
Subjects:	Q Science Q Science > QA Mathematics Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Benedictus Kenny Tjahjono
Date Deposited:	12 Feb 2024 03:10
Last Modified:	12 Feb 2024 03:10
URI:	http://repository.its.ac.id/id/eprint/106911

Actions (login required)

View Item