/VEATGen

A Variational Autoencoder for Texture Generation

Primary LanguagePythonMIT LicenseMIT

VEATGen

Variational Autoencoder for Texture Generation

By Alexandre Variengien

VEATGen image space

Description

I trained VAEs on 3 datasets extracted from the Describable Textures Dataset (DTD).

  • Two 50x50 images datasets: cracked (43 images from the cracked category of the DTD) and honey (30 images from the honeycombed category)
  • One 100x100 images dataset: honey-large (30 images, same as the honey)

Explore the latent space of these pretrained VAE using an intuitive GUI !

Read the report.pdf file to learn more about this project.

Environment

This project was developed in Python 3. The GUI uses matplotlib interactive features and sklearn to compute PCA.

For the VAE, I used the keras (v. 2.3 or higher, running on tensorflow). The original VAE implementation used can be found here.

Context

This project was developed as part of the Computer Graphics and Digital Images course from the computer science first year of Master at ENS de Lyon.