This repository contains the implementation, trained models, and experimental results for the project:

Comparative Analysis of Generative Models for Artistic Image Synthesis on ArtBench-10

Authors:

• Arthur Sophiatti [2022115599]
• Giovanni Faedo [2025267503]

Developed at the University of Coimbra.

The goal of this project is to compare three major generative modeling paradigms for artistic image synthesis on the ArtBench-10 dataset:

• Variational Autoencoders (VAE)
• Generative Adversarial Networks (DCGAN)
• Diffusion Models (Pixel and Latent)

All models were trained on 32x32 RGB images across ten artistic styles.

Evaluation was performed using:

• Fréchet Inception Distance (FID)
• Kernel Inception Distance (KID)

GenAI_Project/
│
├── TP1-alunos-search-only/
│   └── student_start_pack/
│       └── ArtBench-10_Student_Start_Pack.ipynb   # Main notebook
│
├── configs/      # Hyperparameter configurations (.json)
├── results/      # Evaluation metrics and outputs (.json)
├── history/      # Training logs (.json)
├── Comparative_Analysis_of_Generative_Models_for_Artistic_Image_Synthesis_on_ArtBench_10.pdf
└── README.md