Advanced 3D Artistic Image Generation with VAE-SDFCycleGAN

Generation of a 3-dimensional (3D)-based artistic image from a 2-dimensional (2D) image using a generative adversarial network (GAN) framework is challenging.  Most existing artistic GAN-based frameworks lack robust algorithms lack suitable 3D data representations that can fit into    GAN    to prod...

Full description

Saved in:
Bibliographic Details
Main Authors: Dorcas Oladayo Esan, Pius Adewale Owolawi, Chunling Tu
Format: Article
Language:English
Published: Informatics Department, Faculty of Computer Science Bina Darma University 2024-12-01
Series:Journal of Information Systems and Informatics
Subjects:
Online Access:https://journal-isi.org/index.php/isi/article/view/900
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850023685596905472
author Dorcas Oladayo Esan
Pius Adewale Owolawi
Chunling Tu
author_facet Dorcas Oladayo Esan
Pius Adewale Owolawi
Chunling Tu
author_sort Dorcas Oladayo Esan
collection DOAJ
description Generation of a 3-dimensional (3D)-based artistic image from a 2-dimensional (2D) image using a generative adversarial network (GAN) framework is challenging.  Most existing artistic GAN-based frameworks lack robust algorithms lack suitable 3D data representations that can fit into    GAN    to produce high-quality 3D artistic images. To produce 3D artistic images from 2D image that considerably improves scalability and visual quality, this research integrates innovative variational autoencoder signed distance function, cycle generative adversarial network (VAE-SDFCycleGAN). The proposed method feeds a single 2D image into the network to produce a mesh-based 3D shape. The network encodes a 2D image of the 3D object into latent representations, and implicit surface representations of 3D images corresponding to those of 2D images are subsequently generated. VAE extracts feature from the two-dimensional input image and reconstructs a voxel-type grid using a signed distance function. Cycle GAN produces improved and high-quality 3D artistic images from 2D images. The publicly available COCO dataset was used to evaluate the proposed advanced 3D-VAE-SDFCycleGAN. The model produced a peak signal noise ratio (PSNR) of 31.35, mean square error (MSE) of 65.32, and structural similarity index measure (SSIM) of 0.772 which indicates the improved quality of the generated images. The results are compared with other traditional GAN methods and the results obtained show that the proposed method outperforms the others in terms of quantitative and qualitative evaluation metrics.
format Article
id doaj-art-0f28086a19ea4ef4b58f458f796a6416
institution DOAJ
issn 2656-5935
2656-4882
language English
publishDate 2024-12-01
publisher Informatics Department, Faculty of Computer Science Bina Darma University
record_format Article
series Journal of Information Systems and Informatics
spelling doaj-art-0f28086a19ea4ef4b58f458f796a64162025-08-20T03:01:18ZengInformatics Department, Faculty of Computer Science Bina Darma UniversityJournal of Information Systems and Informatics2656-59352656-48822024-12-01642508252410.51519/journalisi.v6i4.900900Advanced 3D Artistic Image Generation with VAE-SDFCycleGANDorcas Oladayo Esan0Pius Adewale Owolawi1Chunling Tu2Tshwane University of Technology South AfricaTshwane University of Technology South AfricaTshwane University of Technology South AfricaGeneration of a 3-dimensional (3D)-based artistic image from a 2-dimensional (2D) image using a generative adversarial network (GAN) framework is challenging.  Most existing artistic GAN-based frameworks lack robust algorithms lack suitable 3D data representations that can fit into    GAN    to produce high-quality 3D artistic images. To produce 3D artistic images from 2D image that considerably improves scalability and visual quality, this research integrates innovative variational autoencoder signed distance function, cycle generative adversarial network (VAE-SDFCycleGAN). The proposed method feeds a single 2D image into the network to produce a mesh-based 3D shape. The network encodes a 2D image of the 3D object into latent representations, and implicit surface representations of 3D images corresponding to those of 2D images are subsequently generated. VAE extracts feature from the two-dimensional input image and reconstructs a voxel-type grid using a signed distance function. Cycle GAN produces improved and high-quality 3D artistic images from 2D images. The publicly available COCO dataset was used to evaluate the proposed advanced 3D-VAE-SDFCycleGAN. The model produced a peak signal noise ratio (PSNR) of 31.35, mean square error (MSE) of 65.32, and structural similarity index measure (SSIM) of 0.772 which indicates the improved quality of the generated images. The results are compared with other traditional GAN methods and the results obtained show that the proposed method outperforms the others in terms of quantitative and qualitative evaluation metrics.https://journal-isi.org/index.php/isi/article/view/9003d image, variational autoencoder, cycle gan, artistic image, signed distance function.
spellingShingle Dorcas Oladayo Esan
Pius Adewale Owolawi
Chunling Tu
Advanced 3D Artistic Image Generation with VAE-SDFCycleGAN
Journal of Information Systems and Informatics
3d image, variational autoencoder, cycle gan, artistic image, signed distance function.
title Advanced 3D Artistic Image Generation with VAE-SDFCycleGAN
title_full Advanced 3D Artistic Image Generation with VAE-SDFCycleGAN
title_fullStr Advanced 3D Artistic Image Generation with VAE-SDFCycleGAN
title_full_unstemmed Advanced 3D Artistic Image Generation with VAE-SDFCycleGAN
title_short Advanced 3D Artistic Image Generation with VAE-SDFCycleGAN
title_sort advanced 3d artistic image generation with vae sdfcyclegan
topic 3d image, variational autoencoder, cycle gan, artistic image, signed distance function.
url https://journal-isi.org/index.php/isi/article/view/900
work_keys_str_mv AT dorcasoladayoesan advanced3dartisticimagegenerationwithvaesdfcyclegan
AT piusadewaleowolawi advanced3dartisticimagegenerationwithvaesdfcyclegan
AT chunlingtu advanced3dartisticimagegenerationwithvaesdfcyclegan