Napovedovanje obrazov staršev na podlagi obrazov njihovih otrok

Potočnik, Vid (author), Šajn, Luka (mentor)

V tej diplomski nalogi bomo najprej opisali problem napovedovanja in generiranja slik obrazov staršev na podlagi slik otrok. Na kratko bomo predstavili razvite metode ChildGAN, KinshipGAN in DNA-Net, ki delujejo na podobnih principih, vendar so razvite za problem napovedovanja in generiranja slik obrazov otrok na podlagi slik obrazov njihovih staršev. Razvili bomo lastno metodo napovedovanja obrazov staršev na podlagi skupne implementacije. Dodali smo novo zbirko podatkov PPLDB, ki dopolnjuje NOKDB, tako da omogoča dodajanje poljubnih slik otrok, katerim lahko napovemo obraze staršev.

• Language: Slovenian
• Keywords: napovedovanje slik obrazov, GAN, StyleGAN, PyTorch, ChildGAN, KinshipGAN, DNA-Net, Families In The Wild, linearna regresija, napovedni modeli, genetika
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2023
• PID: 20.500.12556/RUL-152718
• COBISS.SI-ID: 165501955
• Publication date in RUL: 04.12.2023

Secondary language

• Language: English
• Title: Predicting images of parents’ faces based on images of childrens’ faces

Abstract

In this thesis, we will first describe the problem of predicting and generating images of parents' faces from images of their children. We will briefly introduce the already developed methods ChildGAN, KinshipGAN and DNA-Net, which work on similar principles but are developed for the problem of predicting and generating images of childrens' faces based on images of parents' faces. We will develop our own method for predicting parents' faces based on a common implementation. We have added a new database, PPLDB, which complements NOKDB by allowing to add arbitrary child images to which we can predict faces of their parents.

• Keywords: face image prediction, GAN, StyleGAN, PyTorch, ChildGAN, KinshipGAN, DNA-Net, Families In The Wild, linear regression, prediction models, genetics