Generiranje slika lica pomoću Generativne suparničke mreže

undergraduate thesis

• Title (english): Generating Face Images With Generative Adversarial Network
• Author: Ivan Pavlović
• Mentor: Tomislav Pribanić (mentor)
• Committee member: Marko Subašić (predsjednik povjerenstva)
• Committee member: Donik Vršnak (član povjerenstva)
• Committee member: Ivana Matovinović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-01, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U ovom radu istraženi su modeli za generaciju realističnih slika lica, s naglaskom na usporedbu DCGAN (Deep Convolutional GAN) i WGAN-GP (Wasserstein GAN s gradijentnom kaznom). Implementacija DCGAN-a pokazala je ograničenu sposobnost modela u generiranju raznovrsnih i uvjerljivih slika, često s artefaktima i deformacijama. Primjenom WGAN-GP modela, postignuta su značajna poboljšanja u kvaliteti, varijabilnosti i realizmu generiranih slika. Iz rezultata je vidljivo da WGAN-GP pruža stabilniji proces treniranja i superiornu sposobnost simulacije stvarnih uzoraka. Ovaj rad naglašava važnost odabira arhitekture i trenirajućih tehnika u generaciji visoko kvalitetnih sintetičkih podataka.

Abstract (english)

This paper explores models for generating realistic facial images, with a focus on comparing DCGAN (Deep Convolutional GAN) and WGAN-GP (Wasserstein GAN with gradient penalty). The implementation of DCGAN showed the limited capability of the model in generating diverse and convincing images, often with artifacts and deformations. By applying the WGAN-GP model, significant improvements in the quality, variability, and realism of the generated images were achieved. The results indicate that WGAN-GP provides a more stable training process and superior ability to simulate real-world patterns. This paper emphasizes the importance of choosing architectures and training techniques in generating high-quality synthetic data.

• Keywords: wgan; gan; generativne; suparničke; mreže; lica
• Keywords (english): wgan; gan; generative; adversarial; networks; faces
• Language: croatian
• URN:NBN: urn:nbn:hr:168:554321
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2024-10-12 21:48:04