INTELIGÊNCIA ARTIFICIAL NA GERAÇÃO DE IMAGENS REALISTAS: TÉCNICAS E APLICAÇÕES

Authors

  • Kalil Garcia Canuto Unversidade Federal do Tocantins
  • Warley Gramacho da Silva Universidade Federal do Tocantins

DOI:

https://doi.org/10.20873/vol.13n.3pibic202517

Keywords:

Artificial intelligence, DCGAN, Deep Learning

Abstract

This project investigates the application of Deep Convolutional Generative Adversarial Networks (DCGANs) for the generation of realistic synthetic images. Utilizing the 102 Flower Category Dataset, a model was implemented and trained with the objective of generating 128x128 pixel color images of flowers. The model architecture was based on the principles proposed by Radford, Metz, and Chintala (2016), featuring a Generator that uses transposed convolutions and a Discriminator that employs strided convolutions. The training was stabilized through Batch Normalization and ReLU/LeakyReLU activation functions. The quality of the generated samples was evaluated both visually and quantitatively using the Fréchet Inception Distance (FID) metric. The results demonstrate the architecture's ability to learn the distribution of a complex dataset, achieving an optimal FID score of 808.74 at epoch 600. This result serves as a quantitative baseline and shows the difficulties of using this architecture to generate highly realistic images with this dataset.

References

GOODFELLOW, Ian J. et al. Generative adversarial networks. 2014. Disponível em: https://arxiv.org/abs/1406.2661. Acesso em: 4 set. 2025.

GOODFELLOW, Ian; BENGIO, Yoshua; COURVILLE, Aaron. Deep learning. Cambridge: MIT Press, 2016. Disponível em: http://www.deeplearningbook.org. Acesso em: 4 set. 2025.

HEUSEL, Martin; RAMSAUER, Hubert; UNTERTHINER, Thomas; NESSLER, Bernhard; HOCHREITER, Sepp. GANs trained by a two time-scale update rule converge to a local Nash equilibrium. 2018. Disponível em: https://arxiv.org/abs/1706.08500. Acesso em: 30 ago. 2025.

IOFFE, Sergey; SZEGEDY, Christian. Batch normalization: accelerating deep network training by reducing internal covariate shift. 2015. Disponível em: https://arxiv.org/abs/1502.03167. Acesso em: 22 ago. 2025.

JAYASUMANA, Sadeep et al. Rethinking FID: towards a better evaluation metric for image generation. 2024. Disponível em: https://arxiv.org/abs/2401.09603. Acesso em: 30 ago. 2025.

KARRAS, Tero; AITTALA, Miika; HELLSTEN, Janne; LAINE, Samuli; LEHTINEN, Jaakko; AILA, Timo. Training generative adversarial networks with limited data. 2020. Disponível em: https://arxiv.org/abs/2006.06676. Acesso em: 4 set. 2025.

KINGMA, Diederik P.; BA, Jimmy. Adam: a method for stochastic optimization. 2017. Disponível em: https://arxiv.org/abs/1412.6980. Acesso em: 30 ago. 2025.

RADFORD, Alec; METZ, Luke; CHINTALA, Soumith. Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. 2015. Disponível em: https://arxiv.org/abs/1511.06434. Acesso em: 22 ago. 2025

WAN, Zhijing; WANG, Zhixiang; CHUNG, CheukTing; WANG, Zheng. A survey of dataset refinement for problems in computer vision datasets. 2023. Disponível em: https://arxiv.org/abs/2210.11717. Acesso em: 30 ago. 2025.

Published

2026-05-13

How to Cite

Garcia Canuto, K., & Gramacho da Silva, W. (2026). INTELIGÊNCIA ARTIFICIAL NA GERAÇÃO DE IMAGENS REALISTAS: TÉCNICAS E APLICAÇÕES. DESAFIOS - Revista Interdisciplinar Da Universidade Federal Do Tocantins, 13(3), 197–209. https://doi.org/10.20873/vol.13n.3pibic202517

Issue

Section

PIBIC 2024-2025

Most read articles by the same author(s)

Similar Articles

<< < 1 2 3 4 > >> 

You may also start an advanced similarity search for this article.