DIFMI

Generative models experiments for generating a new Miguel (joabreu).

Setup

poetry install

Models

• VAE with DNN-based Encoder/Decoder: src.models.vae.LinearModel
• VAE with CNN-based Encoder/Decoder: src.models.vae.CNNModel

All experiments are available in Jupyter Notebook difmi.ipynb.

Notes

A single image was used for training, image augmentation using torchvision was performed to enhance results.

Results

CNN-based VAE

Linear-based VAE